1. Field of the Invention
The invention pertains to the field of video transmissions. More particularly, the invention pertains to a system and a method for generating video frames.
2. Description of the Related Art
Virtually all applications of video and visual communication deal with large quantities of video data. To create a video presentation, a rendering computer displays a plurality of digital images (“frames”) in succession, thereby simulating movement.
Currently, certain technical problems exist relating to transmitting and rendering a video presentation across low bandwidth computer networks
One problem that is encountered in current streaming systems is that the transmission bandwidth between the media server 102 and the rendering computer 106 is not sufficient to support a real-time seamless presentation, such as is provided by a standard television set. To overcome this problem and allow the user to receive the presentation in real-time, the video presentation is often spatially and temporally compressed. Further, to reduce the amount of data that is transmitted, the media server 102 skips selected frames of the presentation, or, alternatively, the video presentation can be developed having only a few frames per second. The resulting presentations, however, are jittery and strobe-like and are simply not as smooth as a presentation that has a higher frame rate.
To increase the rate at which the frames are displayed to a user, a frame generator 112 may be used to provide intermediate frames between two selected reference frames of the video presentation 110. Typically, frame generators fall within one of two categories: linear motion interpolation systems and motion compensated frame interpolation systems. Linear motion interpolation systems superimpose two reference frames of the video presentation 110 to create one or more intermediate frames. Motion compensated frame interpolation systems use motion vectors for frame interpolation.
In motion compensated interpolation systems, selected macroblocks are assigned a motion vector based upon a reference frame.
By sending motion vectors that point to regions in the reference frame already transmitted to the rendering computer 106, the media server 102 can transmit a representation of a frame using less data than if the pixel information for each pixel in each block is transmitted.
Although current frame generators increase the frame rate, they are simplistic in design. These systems do not account for certain idiosyncrasies within selected streaming presentations. For example, current frame generators that use motion compensated frame interpolation do not account for video presentations that have textual characters. Often a video image is overlaid with video text to convey additional information to the viewer. If motion compensated frame interpolation generates an intermediate frame having textual characters, the generated frame may inappropriately move the text to a new position, thereby creating some floating text that was not intended by the creator of the video presentation.
Another problem associated with existing frame generators is that they unintelligently perform frame generation regardless of whether such interpolation results in a better quality video presentation. Although frame interpolation does increase the number of frames presented to the viewer, such frame generation can produce strange results under certain circumstances. Some encoders, for example, choose a motion vector for a selected block based only upon the fact that the motion vector references a block that is a good match for the selected block even though there is no actual motion from one corresponding frame to the other. Thus, since all of the vectors do not represent motion, frame generation in these instances should not always be employed.
Additionally, current frame generators do not perform any type of post filtering to the generated frames. As can be readily appreciated, since motion compensated interpolation systems build an intermediate frame using blocks of pixels, i.e., macroblocks, the pixels at the border of each block may not be a close match to the pixels in the neighboring block. Accordingly, the borders of each of the blocks may be readily visible to a viewer of the media presentation.
There is a need for a frame generator that behaves intelligently about the frame generation process. If frame generation would produce anomalous results, frame generation should not be performed. A frame generator should also determine whether the reference frames include textual characters and account for them in the frame generation process. A frame generator should also filter interpolation artifacts from the intermediate frame.
The frame generator of the present invention has several features, no single one of which is solely responsible for its desirable attributes. Without limiting the scope of this invention as expressed by the claims which follow, its more prominent features will now be discussed briefly. After considering this discussion, and particularly after reading the section entitled “Detailed Description of the Invention” one will understand how the features of this invention provide advantages over other frame generators.
One embodiment of the invention includes a method of generating video frames, the method comprising receiving a first frame in a memory in the computer system, the first frame representative of an image at a first instance in time, the first frame including a plurality of first elements and a plurality of motion vectors, each of the motion vectors being associated with one of the plurality of first elements, receiving a second frame in a memory in the computer system, the second frame representative of an image at a second instance in time, the second frame including a plurality of second elements and a plurality of motion vectors, each of the motion vectors being associated with one of the plurality of second elements, modifying at least one of the motion vectors in the first and/or second frames based upon the value of one of the other motion vectors in the first and/or the second frame, and generating at least one intermediate frame based upon the first and/or second elements and the associated motion vectors.
Another embodiment of the invention includes a system for generating frames, the system comprising means for receiving a first frame in a memory in the computer system, the first frame representative of an image at a first instance in time, the first frame including a plurality of first elements and a plurality of motion vectors, each of the motion vectors being associated with one of the plurality of first elements, means for receiving a second frame in a memory in the computer system, the second frame representative of an image at a second instance in time, the second frame including a plurality of second elements and a plurality of motion vectors, each of the motion vectors being associated with one of the plurality of second elements, means for modifying at least one of the motion vectors in the first and/or second frames based upon the value of one of the other motion vectors in the first and/or second frame, and means for generating at least one intermediate frame based upon the first and/or second elements and the associated motion vectors.
Yet another embodiment of the invention includes a system for generating frames, comprising a frame analysis module for receiving frames, each of the frames having a plurality of elements, one or more of the elements having an associated motion vector that identifies a base element, the frame analysis module identifying at least one candidate motion vector, the frame analysis module determining with respect to each of the plurality of elements whether the at least one candidate motion vector identifies a base element which provides a better match than the base element referenced by the motion vector currently associated with a respective element, and if the candidate motion vector identifies a better matching base element, the frame analysis module replacing the motion vector currently associated with the respective element with the candidate motion vector, and a frame synthesis module for generating frames based upon the received frames.
Yet another embodiment of the invention includes a system for generating frames, comprising a frame analysis module for receiving frames, each of the frames having a plurality of elements, one or more of the elements having an associated motion vector that identifies a base element, the frame analysis module determining, with respect to each of the plurality of elements, a median value vector for a group of motion vectors being associated with elements that are positioned proximate to a respective one of the elements, the frame analysis module assigning the median value vector to the respective element, and a frame synthesis module for generating frames based upon the received frames.
Yet another embodiment of the invention includes a method of generating frames, the method comprising receiving a first frame in a memory in the computer system, the first frame representative of an image at a first instance in time, the first frame including a plurality of first elements and a plurality of motion vectors, each of the motion vectors being associated with one of the plurality of first elements, receiving a second frame in a memory in the computer system, the second frame representative of an image at a second instance in time, the second frame including a plurality of second elements and a plurality of motion vectors, each of the motion vectors being associated with one of the plurality of second elements, modifying each of the motion vectors associated with each of the elements based upon the value of at least one of the other motion vectors, and generating at least one intermediate frame based upon the first and second elements and the associated motion vectors.
The following detailed description is directed to certain specific embodiments of the invention. However, the invention can be embodied in a multitude of different ways as defined and covered by the claims. In this description, reference is made to the drawings wherein like parts are designated with like numerals throughout.
It is noted that the media server 102 and the rendering computer 106 may each be any conventional general purpose computer using one or more microprocessors, such as a Pentium processor, a Pentium II processor, a Pentium Pro processor, an xx86 processor, an 8051 processor, a MIPS processor, a Power PC processor, or an ALPHA processor.
The media server 102 includes an encoder 116 for encoding video images. The rendering computer 106 includes a decoder 118 for decoding the video images that have been encoded by the encoder 116 and subsequently transmitted to the rendering computer 106. For convenience of description, the following description will describe the use of an encoder and decoder that are configured to meet the H.263 and/or the H.263(+) Version 2 standards. However, the processes of the present invention may used with other standard encoding and decoding schemes, such as, for example, H.261, MPEG-1, and MPEG-2. Further, a proprietary encoding and decoding scheme may be used.
A frame generator 112 receives from the decoder 118 two frames at a time, each of the frames respectively acting as a first base frame and a second base frame which may be used for the interpolation of one or more intermediate frames.
The frame generator 112, as part of the frame generation process, uses and/or calculates a number of thresholds, each of which are discussed below. Each of the values of these thresholds has been derived analytically and using heuristical testing Accordingly, each of the values can vary depending on the transmission rate of the network 104, the processing speed of the media server 102 and the rendering computer 106, and the encoding and decoding algorithms that are respectively employed by the encoder 116 and the decoder 118.
In one embodiment, the frame generator 112 operates using a state machine 404. In one embodiment the state machine 404 is a software module, similar in kind to the frame analysis module and/or the frame synthesis module 402. The state machine 404 is operably connected to the frame analysis module 400 and the frame synthesis module 402. In one embodiment of the invention, the state machine 404 has seven states, including: KEYFRAME, ABORT, LOW_ACTION, MODERATE_ACTION, HIGH_ACTION, BIG_CHANGE, and SMALL_CHANGE. However, as can be readily appreciated by one of ordinary skill in the art, a simpler or a more complex state machine can be employed.
The frame generator 112 uses one or more states of the state machine 404 to determine which actions need be performed during frame analysis and frame generation. Table 1 provides a brief description of each of the states.
The significance of each of the states of the state machine 404 in relation to the frame generation process will be discussed in further detail below.
Referring again to
Alternatively, the frame analysis module 400, the frame synthesis module 402, and/or the state machine 404 may each be implemented as a hardware device.
Next, at a decision step 504, the frame generator 112 (
If the frame generator 112 determines it is acceptable to generate intermediate frames, the frame generator 112 proceeds to the step 506. At the step 506, the frame generator 112, depending on the implementation, generates one or more intermediate frames. The process for generating intermediate frames is described in further detail below with reference to
Referring again to the decision step 504, if the frame generator 112 (
Starting at a step 600, the decoder 118 (
In an embodiment of the invention using the H.263 standard, the statistical information that is transmitted from the decoder 118 (
At a next step 602, the frame generator 112 (
If the second base frame was independently coded, the frame generator 112 (
If the time between the first and second base frames is greater than the first threshold, the frame generator 112 (
Referring again to the decision step 612, if the frame generator 112 (
Now, referring again to the decision step 604 (
If the duration between the first base frame and second base frame is greater than the second threshold, the frame generator 112 (
Next, at a decision step 624, the frame generator 112 (
If the number of macroblocks in the second base frame that have been independently coded is greater than the third threshold, the frame generator 112 proceeds to the step 614. Otherwise, if the number of macroblocks in the second frame that have been independently coded is less than the third threshold, the frame generator 112 proceeds to a decision step 626.
At a decision step 626, the frame generator 112 (
In one embodiment of the invention, the value of the fourth threshold depends on the value of the quantizer for the second frame. The frame generator 112 uses the quantizer to determine whether any of the motion vectors in the second frame is relatively large in relation to the quantizer. Table 2 illustrates the values of the fourth threshold depending on the value of the quantizer.
If the scalar associated with the maximum motion vector is less than the fourth threshold, the frame generator 112 proceeds to a step 630. At a step 630, the frame generator 112 assigns the state in the state machine 404 (
Referring again to the step decision step 626, if the frame generator 112 (
For example,
Continuing the example, to calculate the difference between the motion vector 908 and the motion vector 918, the following calculations are performed. First, the absolute value of the differences in the x components of each vector is determined, i.e., 2−0=2. Second, the absolute value of the differences in the y components of each component is determined, i.e., 2−2=0.
Next, at a step 634 of
Once the frames are divided into sections, the frame generator 112 (
Proceeding to a step 640 (
Continuing to a step 642, if any of the sums exceeds the fifth threshold, the frame generator 112 proceeds to a step 644 and assigns the state in the state machine 404 (
Referring again to the step 642 (
For example, referring again to
Proceeding to a step 802 (
At a next step 804, the frame generator 112 calculates the average sum-squared difference of pixel intensity for each of the macroblocks. The average sum-squared difference is calculated by summing the average sum-squared difference for each macroblock (calculated in the step 802) and dividing the sum by the number of macroblocks.
Continuing to a decision step 808, the frame generator 112 determines whether the average sum-squared pixel intensity is greater than a sixth threshold. In one embodiment of the invention, the sixth threshold is calculated by Equation 1.
sixth threshold=quantizer*4*(133/Δ)2. (1)
It is noted that the value of 133 corresponds to the time in milliseconds between two frames that are sent at a rate of 7.5 frames per second (a typical transmission rate for a 28.8 kbps communications device).
If the average sum-squared pixel intensity is greater than the sixth threshold, the frame generator 112 (
Referring again to the decision step 808 (
If the sum-squared differences in pixel intensity is less than the seventh threshold, the frame generator 112 (
Referring again to the decision step 814, if the frame generator 112 (
Depending on the state of the state machine 404, the frame generator 112 (
Next, at a decision step 1016, the frame generator 112 (
Referring again to the decision step 1016 (
At the step 1102 (
Continuing to a decision step 1106, if the frame generator 112 (
Referring again to the step 1000 (
Next, at a step 1044, the frame generator 1044 performs motion vector filtering. The process for motion vector filtering is described below with reference to
Next, at a state 1046, the frame generator 112 re-determines the state of the frame generator 112. In one embodiment of the invention, the frame generator 112 re-executes each of the steps shown in
Continuing to a decision step 1048, the frame generator 112 determines whether the state changed from “MODERATE_ACTION” to “HIGH_ACTION” subsequent to executing the step 1046. If the state changed from “MODERATE_ACTION” to “HIGH_ACTION”, the frame generator proceeds to a state 1050. At the state 1050, the frame generator 112 changes the state back to “MODERATE_ACTION.” The frame generator 112 then proceeds to a state 1052.
Referring again to the decision state 1048, if the frame generator 112 determines that the state did not change from “MODERATE_ACTION” to “HIGH_ACTION”, the frame generator proceeds to the state 1052.
From either the decision step 1048 or the step 1050, the frame generator 112 proceeds to a step 1052. At the step 1052, the frame generator 112 repeats the process for motion vector filtering (the process first being performed in step 1044). The process for motion vector filtering is described below with reference to
Next, the frame generator 112 returns to the step 1012 and performs text detection (discussed above).
Referring again to the step 1000 (
Starting at a step 1200, the frame generator 112 (
At a next step 1202, the frame generator 112 (
Continuing to a step 1204, the frame generator 112 (
Once the frame generator 112 (
For example, in regard to block 1302 and referring to
To determine the closeness of a match between the corner block in the second base frame and one of the selected blocks in the first base frame, the frame generator 112 performs Equation 2.
MATCH=SUM+(8192*RADIUS) (2)
The lower the value of MATCH, the better is the match between the selected block in the first base frame and the selected one of the corner blocks 1302, 1306, 1308, 1310.
Continuing to a step 1208, once the blocks in the first base frame which provide a good match for each of the corner blocks 1302, 1304, 106, 1308 are identified, a motion vector is calculated between the corner blocks 1302, 1304, 1306, 1308 and the identified blocks.
The frame generator 112 (
In summary, the group of candidate motion vectors includes: a null vector, the current motion vector, an average motion vector for the second base frame, and the four motion vectors calculated with respect to each of the corner blocks in the steps 1204 and 1208.
Next, in a step 1210, the frame generator 112 determines, for each macroblock in the second base frame, which of the candidate motion vectors should be used for frame generation. To identify the most appropriate motion vector, the frame generator 112 applies Equation 3 in conjunction with a selected macroblock and in sequence with each of the candidate motion vectors.
MATCH=SUM+4096(|x|+|y|). (3)
After applying Equation 3 to each of the candidate motion vectors for a selected macro block, the frame generator 112 examines the value of the variable MATCH that has been determined using each of the candidate motion vectors. The frame generator 112 then selects the motion vector that has the lowest determined MATCH value and resets the motion vector of the macroblock to the selected motion vector. The frame generator 112 then applies Equation 3 with respect to the remainder of the macroblocks and in conjunction with each of the candidate motion vectors.
After starting at a step 1600 (
Moving to a decision step 1606, the frame generator 112 (
Referring again to the step 1606, if the frame generator 112 (
Referring to
From either the step 1610 or the step 1614, the frame generator 112 proceeds to a step 1618. At the step 1618, the frame generator 112 (
Referring to
In one embodiment of the invention, as a part of frame generation, the frame generator 112 marks macroblock quadrants (8×8 blocks of pixels) in the first base frame and the second base frame that are to remain stationary In general, the text detection process follows three steps. First, the frame generator finds putative blocks of text. Second, the frame generator 112 disregards any putative text blocks that are non-contiguous with another text block. Lastly, the frame generator marks as putative text blocks any gaps and corners that are identified in the selected frame.
Starting the text detection process at a step 1900 (
If the frame generator 112 determines that the value calculated for a block in the step 1900 is below the threshold, the frame generator 112 proceeds to a step 1906. At a step 1906, the frame generator 112 reassigns the motion vector of the blocks that fall within the eighth threshold to be equal to 0, 0. The frame generator 112 assumes that if the difference in the pixel intensity of two blocks having the same spatial position in two different frames is negligible, the block contains text and its motion vector for frame generation purposes should be null.
From either the step 1906, or the decision step 1902, assuming the sum of the squared differences for a given blocks was below the eighth threshold, the frame generator 112 (
Otherwise, if the frame generator 112 determines that any of the blocks have not been determined to be text blocks when analyzing the previous two base frames, the frame generator 112 proceeds to a step 1918 with respect to those blocks. At the step 1918, the frame generator 112 sets the value of the variable “threshold” equal to 1600 for each of the blocks that have been determined to be a text block in the previous two frames.
From either the step 1914 or the step 1918, the frame generator 112 proceeds to the step 1922. At the step 1922, the frame generator 112 determines for each block whether adjacent pixels in the block have varying intensity with respect to each other. If two pixels vary sufficiently in intensity, the frame generator 112 identifies these two pixels as forming an “edge” in the image. If enough edges are found in the block, the block is assumed to be a text block.
In one embodiment of the invention, the frame generator 112 applies Equations 4 and 6 with respect to each of the pixels in a selected block in the first base frame except the bottom-most row and the right-most column of pixels. Further, the frame generator 112 applies Equation 5 and 7 with respect to the pixels in the second base frame having the same spatial positioning as the selected pixels in the first base frame.
Eax=sign(A(x,y)−A(x+1,y))×(A(x,y)−A(x+1,y))2. (4)
Ebx=sign(B(x,y)−B(x+1,y))×(B(x,y)−B(x+1,y))2. (5)
Eay=sign(A(x,y)−A(x,y+1))×(A(x,y)−A(x,y+1))2. (6)
Eby=sign(B(x,y)−B(x,y+1))×(B(x,y)−B(x,y+1))2. (7)
The results of Equations 4 and 5 reveal the relative difference in intensity between two horizontally adjacent pixels in each of the first and second base frames. Similarly, the results of Equation 6 and 7 reveal the relative difference in intensity between two vertically adjacent pixels in each of the first and second base frames. For example, if the value of the variable “Eax” is positive, the left-most of the two referenced pixels has the most intensity of the two. Conversely, if the value of the variable “Eax” is negative, the right-most of the two pixels has the most intensity of the two. A similar relationship exists with respect to the pixels that are applied with the variables “Ebx”, “Eay”, and “Eby.” Using the values of the variables “Eax”, “Ebx”, “Eay”, “Eby” and “threshold”, the frame generator 112 can count the number of edges that are formed by each of the pixels within a selected block.
In one embodiment of the present invention, the frame generator 112 uses a variable “edge_count” to assist in determining the number of edges in the selected block. Once the values of “Eax,” “Ebx”, “Eay”, and “Eby” have been calculated, the frame generator 112 adjusts the value of edge_count based upon these values. Table 3 sets forth below a plurality of conditions that are applied to determine the number of edges in the selected block. If the condition identified in the first column of the table is satisfied, the frame generator 112 adjusts the value of the variable edge_count based upon the result identified in the second column of the table
The frame generator 112 applies Equations 4-7 for each of the pixels within the blocks in a selected spatial position in the first and second base frames. Upon analyzing a new block in a different spatial position, the frame generator 112 resets the value of variable “edge_count” equal to zero.
Next, at a step 1926 the frame generator 112 determines which group of blocks are going to be putatively designated as text blocks. In one embodiment of the invention, if the variable “edge_count” is greater than 10 for a selected block, the frame generator 112 designates this block as a putative text block.
Continuing to a step 1930, the frame generator 112 eliminates any block from the putative group of text blocks which is not immediately bordering another one of the putative text blocks. In one embodiment of the invention, the term border refers to the blocks immediately above, below, to the left and to the right of the selected block.
At the step 1930, the frame generator 112 may optionally determine whether there are a sufficient number of putative text blocks in the putative text block group to allow for the conclusion that the putative text blocks represent text. For example, in one embodiment, if the number of putative text blocks is less than 6, the frame generator 112 removes all of the of the blocks from the putative text block group.
Further, at the step 1930, the frame generator 112 may also optionally determine whether the number of blocks in the putative text block group exceeds a text block threshold. If too many text blocks are detected, the frame generation 112 assumes that the first and second base frames contain predominantly text and that the frame generation in this instance may produce unexpected results. If the number of blocks in the putative block exceed the text block threshold, the frame generator 112 fails the frame generation based upon the first and the second base frames. In one embodiment, the text block threshold is predefined to be equal about 30. In another embodiment of the invention, the text block threshold is calculated on an ad-hoc basis.
Continuing to a step 1933, the frame generator 112 analyzes the putative text block groups to determine whether any gaps or missing corners can be identified. In one embodiment of the invention, a selected block is added to the putative text block group if there are putative text blocks: (i) positioned both above and below the selected block; (ii) positioned both to the left and the right of the selected block; (iii) positioned below, to the left, and below and to the left of the selected block; (iv) positioned below, to the right, and below and to the right of the selected block; (v) positioned above, to the left, and above and to the left of the selected block; or (vi) positioned above, to the right, and above and to the right of the selected block.
For example, referring to
In one embodiment of the invention, the frame generator 112 performs a left to right, top to bottom traversal of each of the blocks 2002 in the frame 2000. However, other traversal methods may be employed. Further, in one embodiment of the present invention, the frame generator 112 does not consider newly added putative text blocks when filling in the gaps and missing corners. For example, as was discussed above, block 2032 was added as a putative text block due to the positioning of other putative text blocks. However, in this embodiment, the frame generator 112 does not consider block 2032 as being putatively marked when analyzing the other blocks.
Alternatively, in another embodiment, the frame generator 112 includes newly added putative text blocks when analyzing the remainder of the blocks. For example, in this embodiment, after determining that the block 2032 should be added as a putative text block, the frame generator 112 considers the block 2032 as being one of the putative text blocks for the remainder of the analysis. Further, in this embodiment, the frame generator 112 may perform multiple traversals of the blocks 2002, each traversal filling in new gaps or empty corners that are identified in a previous traversal.
It is noted that the frame generator 112 may be called more than once thereby generating more than one intermediary frames. In one embodiment of the invention, the frame generator 112 receives a time reference that identifies where in time the generated frame is to be displayed in relation to the first and second base frame. For example, a first base frame may be designated for display at a time 100 milliseconds. Further, the second base frame may be designated for display at a time 300 milliseconds. The frame generator 112 using the first and second base frames can generate one intermediate name for display at a time 200 milliseconds. Alternatively, the frame generator 112 can be requested to provide two intermediate frames, one for display at a time 150 milliseconds, the other at a time 250 milliseconds. The frame generator 112 can be used to generate any number of intermediate frames. Further, the frame generator 112 can be used to generate yet another frame using a generated frame as one of the base frames. Also, the frame generator 112 can be adapted to receive a request that identifies one or more intermediate times instead of being provided the intermediate times in succession.
Moreover, for example, the intermediate time can be represented as an integer ranging between the values 1 and 255. In this embodiment, an intermediate time of 1 represents that the frame to be generated is to be presented proximal in time to the first base frame. An intermediate time of 255 represents that the frame to be generated is to be presented proximal in time to the second base frame. For convenience of description, the remainder of the description will describe the operation of frame synthesis with reference to a time interval that is measured as a value falling between 1 and 255.
Continuing to a step 2104, the frame generator 112 determines whether the state of the state machine 404 is equal to “MODERATE_ACTION.” If the step is not equal to “MODERATE_ACTION”, the frame generator 112 proceeds to a step 2108.
At the step 2108, the frame generator 112 merges the first and second base frames using a linear average. In one embodiment of the invention, as part of this step, the frame generator 112 applies Equation 8 for each pixel position within the first and second base frames.
generated frame(x,y)=(256−time reference)/256*first frame(x, y)+time_reference/256*second frame(x, y)
Referring again to the step 2104, if the frame generator 112 determines that the state is equal to “MODERATE_ACTION”, the frame generator 112 proceeds to the step 2112. At the state 2112, the frame generator 112 determines whether the frame to be generated is to be presented in time closer to the first base frame or to the second base frame. If the frame generator 112 determines that the frame to be generated is to be presented closer in time to the second base frame, the frame generator 112 proceeds to a step 2120. Otherwise, if the frame generator 112 determines that the frame to be generated is to be presented closer in time to the first base frame, the frame generator 112 proceeds to a step 2124. For example, in the embodiment of the invention, if the value of the variable “time_reference” is greater than or equal to 128, the frame generator 112 determines that the frame to be generated is closer to the second base frame. Furthermore, if the value of the variable “time_reference” is less than 128, the frame generator 112 determines that the frame to be generated is closer to the first base frame.
At the state 2120, the frame generator 112 generates a frame using the macroblock information from the second base frame. As part of the state 2120, the frame generator 112 generates four “reference” motion vectors for each of the macroblocks from the motion vector of the selected macroblock and motion vectors of neighboring macroblocks. Each motion vector represents motion of one of four quadrants, or “blocks”, within a selected macroblock. If the selected macroblock is located on the frame edge, then motion vectors for those quadrants along the edge are given a motion vector equal to the one for the entire macroblock Otherwise, for quadrants internal to the frame, the frame generator 112 applies Equations 9-16 with respect to each of the macroblock positions in the frame to be generated.
x
upper
left=−((256−time_reference)/256)*((4*MV(x,y).x+2*MV(x,y−1).x+2*MV(x−1,y).x+MV(x−1,y−1).x)/9) (9)
y
upper
left=−((256−time_reference)/256)*((4*MV(x,y).y+2*MV(x,y−1).y+2*MV(x−1,y).y+MV(x−1,y−1).y)/9) (10)
x
lower
left=−((256−time_reference)/256)*((4*MV(x,y).x+2*MV(x,y+1).x+2*MV(x−1,y).x+MV(x−1,y+1).x)/9) (11)
y
lower
left=−((256−time_reference)/256)*((4*MV(x,y).y+2*MV(x,y+1).y+2*MV(x−1,y).y+MV(x−1,y+1).y)/9) (12)
x
upper
right=−((256−time_reference)/256)*((4*MV(x,y).x+2*MV(x,y+1).x+2*MV(x+1,y).x+MV(x−1,y−1).x)/9) (13)
y
upper
right=−((256−time_reference)/256)*((4*MV(x,y).y+2*MV(x,y−1).y+2*MV(x+1,y).y+MV(x+1,y−1).y)/9) (14)
x
lower
right=−((256−time_reference)/256)*((4*MV(x,y).x+2*MV(x,y+1).x+2*MV(x+1,y).x+MV(x+1,y+1).x)/9) (15)
y
lower
right=−((256−time_reference)/256)*((4*MV(x,y).y+2*MV(x,y+1).y+2*MV(x−1,y).y+MV(x+1,y+1).y)/9) (16)
Equations 9-16 collectively yield the x and y components of the four “reference” motion vector. The frame generator 112 uses the reference motion vectors <xupper
Referring again to the decision step 2112, if the frame generator 112 determines that the frame to be generated is to be presented closer in time to the first base frame, the name generator proceeds to the step 2124. At the step 2124, the frame generator 112 generates the intermediate frame using the macroblock information from the second base frame. As part of the state 2124, the frame generator 112 generates four “reference” motion vectors from the motion vector of the selected macroblock and motion vectors of neighboring macroblocks. Each motion vector represents motion of one of four quadrants, or “blocks”, within a selected macroblock. If the selected macroblock is located on the frame edge, then motion vectors for those quadrants along the edge are given a motion vector equal to the one for the entire macroblock. Otherwise, for quadrants internal to the frame, the frame generator 112 applies Equations 17-24 with respect to each of the macroblock positions in the frame to be generated
x
upper
left=(time_reference)/256)*((4*MV(x,y).x+2*MV(x,y−1).x+2*MV(x−1,y).x+MV(x−1,y−1).x)/9) (17)
y
upper
left=(time_reference)/256)*((4*MV(x,y).y+2*MV(x,y−1).y+2*MV(x−1,y).y+MV(x−1,y−1).y)/9) (18)
x
lower
left=(time_reference)/256)*((4*MV(x,y).x+2*MV(x,y+1).x+2*MV(x−1,y).x+MV(x−1,y+1).x)/9) (19)
y
lower
left=(time_reference)/256)*((4*MV(x,y).y+2*MV(x,y+1).y+2*MV(x−1,y).y+MV(x−1,y+1).y)/9) (20)
x
upper
right=(time_reference)/256)*((4*MV(x,y).x+2*MV(x,y−1).x+2*MV(x+1,y).x+MV(x−1,y−1).x)/9) (21)
y
upper
right=(time_reference)/256)*((4*MV(x,y).y+2*MV(x,y−1).y+2*MV(x+1,y).y+MV(x+1,y−1).y)/9) (22)
x
lower
right=(time_reference)/256)*((4*MV(x,y).x+2*MV(x,y+1).x+2*MV(x+1,y).x+MV(x+1,y+1).x)/9) (23)
y
lower
right=(time_reference)/256)*((4*MV(x,y).y+2*MV(x,y+1).y+2*MV(x−1,y).y+MV(x+1,y+1).y)/9) (24)
Equations 17-24 collectively yield the x and y components of the four “reference” motion vector. The frame generator 112 uses the reference motion vectors <xupper
From either the step 2124 or the step 2120, the frame generator 112 proceeds to the step 2128. At the step 2128, for each of the macroblock quadrants having video text in the base frame which was selected in the state 2112, i.e., the first or the second base frame, the frame generator 112 blends each of the pixels in the macroblock quadrant with a macroblock quadrant identically positioned within the other base frame. As part of the blending process, the frame generator 112 applies Equation 8 with respect to each of the pixels in the macroblock quadrants.
Continuing to a state 2128, the frame generator 112 filters the intermediate frame. The process for filtering the intermediate frame is set forth below with reference to
Finally, in the state 2132, the generated intermediate frame is displayed to the user. In one embodiment of the invention, the frame generator displays the intermediate frame to a display on the rendering computer 106 at a time corresponding to the offset time provided to the frame generator 112. In another embodiment of the invention, the intermediate frame is transmitted to a rendering program (not shown).
As part of this process, the frame generator 112 analyzes for each block, the blocks to the right and below the selected block. The frame generator 112 first performs horizontal filtering of each of the rows of each of the blocks. Next, the frame generator 112 performs vertical filtering of each of the rows of each of the blocks. Steps 2200, 2204, 2208, 2212, and 2216 describe the process for horizontal filtering each of the rows in the blocks. Steps 2218, 2219, 2220, 2224 and 2228 describe the process for vertically filtering each of the columns in each of the blocks. In one embodiment of the invention, the frame generator 112 performs two left to right, top to bottom traversals of each of the blocks in the interpolated frame, one traversal for horizontal filtering, the other for vertical filtering. However, it is to be appreciated that other traversal schemes may be used.
Starting at a step 2200, the frame generator 112 begins the horizontal filtering process. The frame generator successively filters each of the rows in each of 8×8 blocks in the interpolated frame. Steps 2204, 2208, 2212, and 2216 further the describe the process for horizontally filtering each of the blocks. Steps 2204, 2208, 2212, and 2216 are executed for each of the 8×8 blocks in the interpolated frames.
Continuing to a step 2204, the frame generator 112 uses the quantizer for the interpolated frame to determine the strength of the filter that is to be used for horizontally filtering the block that is currently selected. In one embodiment of the invention, the quantizer for the interpolated frame is the same value as the quantizer for the base frame that was used to generate the interpolated frame. In one embodiment of the invention, the frame generator 112 uses Equation 25 for this calculation.
filter strength=SF×(|MV(x,y).x−MV(x+1,y)x|+|MV(x,y).y−MV(x+1,y).y|). (25)
The variable “filter_strength” is used by the frame generator 112 to determine the appropriateness of filtering each of the pixels in the currently selected block. It is noted that although the value of the variable “filter_strength” changes during the filtering process for a selected row, the value of the variable “filter_strength” is reset to the filter strength that is associated with the block upon filtering a new row.
Referring now to the steps 2208, 2212, and 2216, the process for horizontally filtering a selected block is described. The frame generator 112 begins an iterative process for each of the rows in the currently selected block. In overview, at the step 2208, the frame generator 112 selects one of the rows in the blocks. Next, at the step 2212, the frame generator 112 determines whether to filter selected pixels in the row. Lastly, in the step 2216, the frame generator 112 filters the selected pixels.
Referring again to the step 2208, the frame generator 112 filters a selected row of the currently selected block. Starting at the top row of the block, the frame generator 112 sets a variable “filter1” to reference the right most pixel of the selected row. Further, the frame generator 112 sets a variable “filter2” to reference the left most pixel in the same row that is in the block to the right of the selected block.
Moving to a step 2212, the frame generator 112 determines whether to filter the pixel referenced by the variable “filter1”, depending on the value of the variable “filter_strength.” In one embodiment of the invention, if the variable “filter_strength” is zero, the frame generator 112 stops filtering the currently selected row. However, it is to be appreciated that other thresholds may be used.
Otherwise, if the value of the variable “filter_strength” is greater than zero, the frame generator 112 proceeds to a step 2216. At the step 2216, the frame generator 112 filters the pixels referenced by the variables “filter1” and “filter2.” The frame generator 112 assigns the pixel referenced by the variable “filter1” to equal the average of the pixel values of itself and of its neighbors to the left and to the right. Further, the frame generator 112 assigns the pixel referenced by the variable “filter2” to equal the average of the pixel values of itself and of its neighbors to the left and to the right. The frame generator 112 then divides the variable “filter_strength” by 2, rounding down. Next, the frame generator 112 reassigns the variable “filter1” to reference the pixel that is immediately to the left of the pixel that is referenced by “filter1.” Further, the frame generator 112 reassigns the variable “filter2” to reference the pixel that is immediately to the right of the pixel that is currently referenced by the variable “filter2.” The frame generator 112 then returns to the step 2212 and continues to filter the pixels in the selected row until the value of the variable “filter strength” is equal to zero At this point, the frame generator 112 returns to the step 2208, to filter the next row in the selected block. Upon completion of filtering each of the rows in the currently selected block, the frame generator 112 returns to the step 2204 to recalculate the value of the variable “filter strength” for the next block which is to be selected.
Once all of the rows of pixels in each block of the interpolated frame have been filtered, the frame generator 112 proceeds to a step 2218 to filter each of the columns of each of the blocks. At a step 2218, the frame generator 112 successively filters each column of pixels in each block of the interpolated frame. The process for determining whether to filter the selected pixels within each of the rows is described in the step 2224 and 2228.
Continuing to a step 2219, the frame generator 112 uses the quantizer for the interpolated frame to determine the strength of the filter that is to be used for vertically filtering each of the blocks in the interpolated frame. In one embodiment of the invention, the frame generator 112 uses Equation 26 for this calculation.
filter_strength=SF×(|MV(x,y).x−MV(x,y+1).x|+|MV(x,y).y−MV(x,y+1).y|). (26)
As was used above in reference to horizontal filtering, the variable “filter_strength” is used by the frame generator 112 to determine the appropriateness of filtering each of the pixels in a selected column of the current block.
Now, referring to steps 2220, 2224, and 2228, the process for vertically filtering the columns of pixels in the blocks is described. At these steps, the frame generator 112 begins an iterative process for each of the columns in a selected one of the blocks. In overview, at the step 2208, the frame generator 112 selects one of the columns in the selected block. Next, at the step 2212, the frame generator 112 determines whether to filter the pixels in the selected column. Lastly, in the step 2216, the frame generator 112 filters the selected pixels.
Referring specifically now to the step 2220, the frame generator 112 filters a selected row of the selected block. Starting at the left most column, the frame generator 112 sets a variable “filter1” to reference the bottom most pixel of the selected column. Further, the frame generator 112 sets a variable “filter2” to reference the top most pixel in the same column that is in the block beneath the selected block.
Moving to a step 2224, the frame generator 112, determines whether to filter the pixel referenced by the variable “filter1”, depending on the value of the variable “filter_strength.” In one embodiment of the invention, if the variable “filter_strength” is equal to zero, the frame generator 112 stops filtering the currently selected column. At this point, if not all of the columns in the block have been selected, the frame generator 112 returns to the step 2220 to filter the next column in the block. Otherwise, if all of the columns in the block have been filtered, the frame generator returns to the step 2218 to filter the next block.
Still referring to thee step 2224, if the variable “filter_strength” is greater than zero, the frame generator 112 proceeds to a step 2228 and filters the pixel referenced by the variables “filter1” and “filter2.” The frame generator 112 assigns the pixel referenced by the variable “filter1” to equal the average of the pixel values itself and of its neighbors to the above and below. Further, the frame generator 112 assigns the pixel referenced by the variable “filter2” to equal the average of the pixel values of itself and of its neighbors to the above and below. The frame generator 112 then divides the variable “filter_strength” by 2, rounding down. Next, the frame generator 112 reassigns the variable “filter1” to reference the pixel that is immediately above of the pixel that is referenced by “filter1.” Further, the frame generator 112 reassigns the variable “filter2” to reference the pixel that is immediately below the pixel that is currently referenced by the variable “filter2.” The frame generator 112 then returns to the step 2224 and continues to filter the pixels in the selected row until the filter strength is equal to zero. At this point, the frame generator 112 returns to the step 2220, to filter the next pixel in the column.
Advantageously, the frame generator 112 of the present invention performs extensive analysis regarding first and second base frames to determine the appropriateness of frame generation. If frame generation would likely cause an anomalous result, frame generation is not performed. Also, if used in conjunction with a motion compensated interpolation system, the frame generator 112 performs global motion vector correction, thereby ensuring that each of the motion vectors represent movement.
Furthermore, the frame generator 112 can detect the presence of text in the first and base frames. If such text is detected, the frame generator maintains the positioning of the text in the generated frames. This is to be contrasted with more simplistic solutions which perform frame generation heedless of the presence of text.
In addition, the frame generator 112 filters the generated frames. Such filtering provides a more pleasing presentation than in presentations having unfiltered frames.
While the above detailed description has shown, described, and pointed out novel features of the invention as applied to various embodiments, it will be understood that various omissions, substitutions, and changes in the form and details of the device or process illustrated may be made by those skilled in the art without departing from the spirit of the invention. The scope of the invention is indicated by the appended claims rather than by the foregoing description. All changes which come within the meaning and range of equivalency of the claims are to be embraced within their scope.
This application is a continuation of U.S. patent application Ser. No. 09/345,577, filed on Jun. 30, 1999, which is hereby incorporated by reference in its entirety. The present application is also related to and incorporates by reference in their entirety the following U.S. patent applications: application Ser. No. 09/345,686, entitled “System and Method for Generating Video Frames,” filed Jun. 30, 1999; application Ser. No. 09/345,576, entitled “System and Method for Generating Video Frames and Detecting Text,” filed Jun. 30, 1999; and application Ser. No. 09/345,584, entitled “System and Method for Generating Video Frames and Post Filtering,” filed Jun. 30, 1999.
Number | Date | Country | |
---|---|---|---|
Parent | 10838991 | May 2004 | US |
Child | 12786342 | US | |
Parent | 09345577 | Jun 1999 | US |
Child | 10838991 | US |