1. Field of the Invention
The present invention relates to a technology that allows shooting a high quality still and/or moving picture. In particular, it is an object of the present invention to provide a computer image processing apparatus for images shot with a camera, which allows generating a clear image free from blur by hand movement in shooting the image and allows shooting a moving picture tracking a target subject by preventing the jitter of an image due to a hand movement by capturing the subject properly.
2. Description of the Related Art
Conventionally, a large amount of moving picture data is reduced by calculating motion vectors between moving picture frames and generating moving picture data with the motion vectors and a reference image.
On the other hand, a technology has been disclosed for correcting a blur in an image due to a hand movement in shooting it by calculating motion vectors between multiple still images shot by an imaging apparatus such as a camera (refer to Patent Document 1, for example).
The reliability of a motion vector must be increased in order to reproduce the moving picture more accurately and/or correcting the blur due to a hand movement in shooting. As a technology for increasing the reliability of a motion vector, Patent Documents 4 to 6 disclose technologies where processing by handling highly correlated motion vectors as those with high reliability is performed.
Furthermore, a technology for maintaining the accuracy of motion vectors and, at the same time, reducing the data amount to calculate more may include identifying feature points of images and calculating motion vectors about the feature points or forming a polygon by connecting the feature points, recognizing how the polygon is deformed by the movement of the feature points among the images to be compared and estimating the positions where the other points are moved (refer to Patent Documents 2 and 3, for example).
Another technology has been disclosed that uses an affine parameter as motion data in order to increase the freedom of the detection of movement such as a rotation of an image and to efficiently perform a correction process with high accuracy (refer to Patent Document 7, for example)
Patent Document 1: Japanese Unexamined Patent Application Publication No. 2003-78807
Patent Document 2: Japanese Unexamined Patent Application Publication No. 11-506576
Patent Document 3: Japanese Unexamined Patent Application Publication No. 10-341445
Patent Document 4: Japanese Unexamined Patent Application Publication No. 9-73540
Patent Document 5: Japanese Unexamined Patent Application Publication No. 6-311502
Patent Document 6: Japanese Unexamined Patent Application Publication No. 6-153146
Patent Document 7: Japanese Unexamined Patent Application Publication No. 9-98424
By the way, when the function for generating motion data such as a motion vector, an affine parameter or the like is implemented by not special hardware but a computer having a calculation function, how a load on the computer is to be reduced during calculating motion data with high accuracy is a problem.
The present invention was made in view of the problem, and it is an object of the present invention to provide an image processing apparatus which can calculate motion data with high precision by simpler processing with a smaller amount of data and lighter load on a computer and allows the correction of image blur due to a hand movement in photo-shooting, generation of high quality photographs by using the technology and capture of a subject properly in moving-picture shooting.
In order to achieve the object, there is provided according to an aspect of the invention, an image processing apparatus including an image data input unit for inputting captured multiple images, a detection image generation unit for generating, based on the input images, smaller images by scaling down a certain area of the images, a multi-resolution image data storage unit for storing multi-resolution image data of the smaller images up to a predetermined resolution level such that each of them is identifiable between a reference image and a target image, an feature point set extraction unit for extracting one or more feature point coordinates based on an image corner or edge of the multi-resolution image data of the reference image, and a motion data calculation unit for calculating motion data between a reference image and the target image by using the multi-resolution image data, wherein the motion data calculating unit has a whole-range matching section for performing, with regard to each of the reference images and target images from a lowest resolution L to a predetermined resolution M (M≦L), a matching process on all image coordinates in a predetermined range of the reference images and calculating motion data based on a first energy function whose arguments are the coordinates on the reference images and the target images, and a motion data calculating section for, with regard to image data of a resolution level higher than the predetermined resolution M, performing a block matching process based on a second energy function by setting input motion data as an initial value and handling the feature point coordinates on the reference images, which are extracted by the feature point set extraction unit, as arguments thereof, and for calculating motion data between the reference images and target images by using reliability defined based on energy of the block matching process or strength of feature points and correlation defined based on motion data among the feature points of the reference images or coordinates based on the motion data, and motion data is calculated by a process, wherein the process comprises (a) calculating motion data for image data of a resolution M−1, which is a higher resolution than the resolution M, with using motion data for image data of the resolution M, which is calculated by the whole-range matching section, as initial value of motion data to be input to the motion data calculating section, (b) calculating motion data for image data of a higher resolution with using the motion data obtained in the previous calculating process as initial value information of motion data to be input to the motion data calculating section, and (c) repeating the calculation for the image data of higher resolution for predetermined number of times or until a predetermined condition is satisfied.
According to the present invention, multi-resolution images of a reference image and one or more target images are generated. A target image is an image to be compared with the reference image. Then, whole-range matching is performed on image data of a lower resolution to detect a two-dimensional displacement between the reference image and each of the target images. Block matching is performed on image data of a higher resolution to detect a displacement at each feature point. The accuracy of motion data is increased by correcting the motion data of an image of a higher resolution by using the previously calculated motion data sequentially from the lowest resolution to higher resolutions as an initial value.
The “specific range” of the detection image generation unit is not limited to a predetermined certain range but may include a range dynamically depending on a condition. Since the main function of the image processing apparatus is the detection of motion data, the name is replaceable by other names such as “motion data detecting apparatus”. The present invention is not limited by the name of the apparatus, and it is intended that the present invention includes all of apparatus having the characteristics above.
Preferably, the motion data to be calculated by the motion data calculating section is an affine parameter. The calculation of the final output at least with an affine parameter allows the detection of movement with a high degree of freedom.
According to the present invention, a two-dimensional (X and Y axes) movement is detected from image data of a lowest resolution or lower resolution than a predetermined resolution by performing whole-range matching, and block matching is used for image data of the predetermined resolution or higher. Furthermore, an affine parameter is used as motion data. Thus, the possibility of the occurrence of a matching error at the lowest resolution can be reduced, and even a movement involving a displaced component such as a rotation can be calculated efficiently.
An image processing apparatus according to the present invention includes an image synthesis unit for calculating an amount of displacement in directions X and Y of two-dimensional coordinates of an input image with reference to a reference image by using the motion data calculated by the motion data calculation unit and based on the degree of displacement between blocks of the target image, dividing the reference image into blocks such that the amount of displacement between blocks can be equal to or lower than a predetermined threshold, calculating two-dimensional travel of the blocks and performing a superposition and/or synthesis process on each block based on the travel.
For example, a superposition process is performed by making approximation of a movement due to a rotation as displacements in X and Y directions, determining the size of blocks for image synthesis and performing block division so as to obtain a displacement of a predetermined or lower value.
In an image processing apparatus according to the present invention, the synthesis unit determines whether coordinate points of four corners of a pre-specified rectangle within a reference image position within a target image or not based on the conversion with motion data, and does not perform superposition of the target image if any one of coordinates positions-outside of the target image or if the amount of displacement between blocks on the coordinates system of the target image is equal to or higher than a predetermined value.
More preferably, the synthesis unit calculates a difference between the reference image and target image in blocks or between a synthesized image and the target image, such as a difference in intensity and/or color of each pixel within the blocks and changes the ratio of superposition to the reference image in blocks based on the difference. The difference may be a sum of square or absolute value of a difference in intensity and/or color. The generation of weighted average image can thin a moving object and clarify a still object only.
According to another aspect of the invention, there is provided an image processing apparatus including an image data input unit for inputting captured multiple images, a detection image generation unit for generating, based on the input images, smaller images of the images in a specific range, a multi-resolution data generation unit for calculating multi-resolution image data up to a predetermined resolution level based on images generated by the detection image generation unit, a frame transforming data generation unit having a motion data detection section for calculating motion data from a reference frame based on the reference frame and an input image, a frame transforming data combining section for calculating a primary frame transforming data based on the motion data and reference frame transforming data, a frame transforming data standardization section for standardizing the primary frame transforming data to calculate secondary frame transforming data, a frame transforming data regularizing section for determining whether a frame transformed by the secondary frame transforming data is within the captured frame or not and regularizing the secondary frame transforming data to make the frame transformed by the secondary frame transforming data fit within the captured frame, and an update section for updating the reference frame and the reference frame transforming data by using the frame transforming data calculated by the frame transforming data regularizing section.
According to the present invention, the reference frame in moving-picture shooting can be updated in real time, and a clear image resulting from the correction of blur due to a hand movement about the reference frame can be generated.
The image processing apparatus may further include a synthesis unit for determining whether coordinate points of four corners of a correction frame rectangle transformed by the frame transforming data are within a shot frame rectangle or not, calculating an optimum frame transforming data to fit the correction frame rectangle within the captured frame rectangle when any of the coordinate points thereof is outside of the captured frame rectangle, and synthesizing the correction frame.
Thus, only the optimum frame transforming data can be output. Maximizing the rectangular area, for example, is one of the methods to obtain the optimum frame transforming data.
According to the invention, two-dimensional motion data is calculated by first performing whole-range matching on image data of a lowest resolution level, and motion data with a high degree of freedom is calculated sequentially by using image data of a higher resolution. Thus, the occurrence of a matching error at the lowest resolution level can be prevented, and highly accurate motion data can be calculated efficiently. A clear image can be generated by synthesizing an image by dividing the image into blocks with a smaller amount of displacement and compensating movements. Thus, a high quality photograph can be generated by reducing image blur due to a hand movement in photo-shooting, and a subject in moving-picture shooting can be properly captured.
Embodiments of the present invention will be described below.
Here, the motion data detection apparatus 10 includes a calculation processing section 13 for calculating motion data such as a motion vector and an affine parameter based on input image data, and a storage section 15 for storing input image data and/or motion data.
The calculation processing section 13 includes an image data input unit (function) 31 for receiving the input of image data from an external device such as a camera, a communication network and a recording medium and storing them in the storage section 15, a detection image generation unit (function) 32 for generating an image (detection image) from which motion data is detected based on input image data, a multi-resolution image generation unit (function) 33 for generating image data of multiple resolution levels by performing a filtering process, a feature point set extraction unit (function) 34 for extracting one or more feature points of an image, a motion data calculation unit (function) 35 for calculating motion data by using multi-resolution image data and/or feature point data, a motion data output unit (function) 36 for outputting calculated motion data. The units 31 to 36 are implemented by functions of a CPU. These units do not have to be separated but may also serve as other units. For example, the motion data calculation unit 35 may also function as the motion data output unit 36.
Here, the data to be stored in the storage section 15 will be described. Image data is data having w×h grids where the width is w and the height is h and having a pixel value (such as RGB and YCbCr) at each of the grids.
Image feature point set data is a set of feature point data of image data, and each feature point includes the coordinate value (x,y) and/or a feature amount indicating the strength of a feature point.
Motion data of an image may be generally an affine parameter or a motion vector. A motion vector may be motion vector mesh data which has multiple motion vector points, motion vector lines and/or motion vector areas.
Next, a method for calculating motion data by operating the apparatus 10 (motion data calculation method) will be described.
[Image Data Input Process]
First, externally input image data is stored in an input image data storage unit 51 by the image data input unit 31.
[Detection Image Generation Process]
Next, the detection image generation unit 32 is started, and an image (detection image), on which the detection of motion data is based, is generated from an input image.
When the detection image generation unit 32 is started, a thumbnail image, which is a smaller input image, is generated, and a clipped image is generated by cutting off the four sides of the image by a predetermined proportion (such as vertically and horizontally 10% off), and the correction of the aspect ratio is performed thereon as required. An intensity image is generated from the corrected image and is stored as a detection image in a detection image data storage unit 52.
[Multi-Resolution Image Generation Process]
Next, the detection image undergoes a filtering process by the multi-resolution image generation unit 33, and the image of a low resolution is generated and is stored in a multi-resolution image data storage unit 53.
The multi-resolution image generation unit 33 upon startup may perform an edge image generation process, for example, on the detection image, and an image of the resolution level 0 is generated. Furthermore, image data of lower resolutions are generated by applying a blur filter and/or averaging filter on each image of the resolution level 0, and the generated multi-resolution image data is stored in the multi-resolution image data storage unit 53. Instead of the use of a blur filter, for example, the multi-resolution image may be scaled down by a certain ratio, and the resolution may be reduced in a stepwise manner.
[Feature Point Set Extraction Process]
Next, the feature point set extraction unit 34 upon startup extracts feature points based on the strength of edges and/or corners. When a feature point is generated at one position, the generation of other feature points is prevented in the vicinity. This is because motion data at one position in an image may have a substantially identical value to that of motion data in the vicinity, and feature points are preferred to be distributed entirely and evenly in order to detect the movement of the entire image.
[Motion Data Calculation Process]
The motion data calculation unit 35 uses a reference image and target image stored in the multi-resolution image data storage unit 53 to calculate motion data by following the steps below:
[Step 1]
Whole-range matching is performed for the roughest resolution (lowest resolution) in multi-resolution matching, and a parallel travel (dx,dy) between two images is obtained. Making the roughest image, at most, several tens pixels×several tens pixels, for example, does not considerably increase the calculation costs for the whole-range matching. Furthermore, the calculation of the parallel travel (dx,dy) is only required because the parallel travel is dominant in the entire movement in hand-movement correction.
Here, I(x,y) is the lightness at coordinates (x,y) of an input image, I′(x,y) is the lightness at the coordinates (x,y) of an output local lightness image, and s is a constant that determines the size of a local area. The local lightness given by the equation represents a relative lightness of a pixel to the average lightness of the pixel and the surroundings(see
Examples of actual input images I and I′ are shown in
This process can correct a change of light condition due to a hand movement and the darkness of an image edge part caused by a low quality imaging lens, for example.
Next, an optimum travel calculation module calculates the travel (dx,dy) from the image I1 to the image I2. First, the travel energy E(dx,dy) of the travel from I1 to I2 by (dx,dy) is defined as:
Here, W indicates the width of an image, and H is the height of the image. E(dx,dy) is the average of the absolute values of differences in pixel values in a overlapping part (shaded part in
As E(dx,dy) decreases, the degree of similarity between images increases. The optimum travel modulation module outputs (dx,dy) such that the value can be minimum.
The process above provides the travel (dx,dy) from the image I1 to the image I2 as motion data.
[Step S2]
Next, the motion data is handled as the initial value of the motion data of an image of the resolution L−1, and a block matching process therewith is performed.
An affine parameter is calculated by the least square method as motion data for images of resolutions L−1 and higher.
The equation in this case is as follows:
The affine parameter calculated by the process is used as the initial value of motion data of the next resolution.
[Step S3]
Next, whether the process reaches the final resolution (which is the resolution level 0 normally) or not is determined. If not, processing in step S2 and subsequent steps are repeated on image data of the next higher resolution level. The motion data of the final resolution is stored in the motion data storage unit 55.
Having described the case that the whole-range matching process is only performed on images of the lowest resolution L in step S1 above, the whole-range matching process may be performed on image data of the lowest resolution L to a certain resolution M (M≦L) with a higher resolution level, and the block matching process in step S2 and subsequent steps may be performed on image data of resolutions from a resolution M−1 with a higher resolution level.
[Motion Data Output Process]
The motion data output unit 36 outputs motion data calculated by the motion data calculation unit 35 and stored in the motion data storage unit 55 to an external storage device, a communication network or another program module, for example.
The motion data detection device 10 can operate on a standalone basis by capturing image data through a storage medium or over a network and also can perform image processing in real time by capturing image data by various equipment such as a camera. For example, the device may be built into a camera module to implement a camera with a hand-movement correction device (or an anti-blur function).
According to this embodiment, multi-resolution images of a reference image and a target image are generated, and whole-range matching is first performed on an image of a lower resolution to detect a two dimensional displacement between the images. A block matching process is performed on an image of a higher resolution to detect a displacement of a high degree of freedom, such as an affine parameter. The accuracy of motion data is increased by calculating the motion data from the lowest resolution to higher resolutions, and calculating the motion data at the next higher resolution by using the previously calculated motion data as an initial value of the motion data at the next higher resolution. Thus, the matching error for lower resolutions can be prevented, and highly accurate motion data can be calculated efficiently. Therefore, the hand-movement correction in shooting with a digital camera can be performed quickly.
Next, operations of the image processing apparatus 1 according to this embodiment will be described. Since the operations of the motion data detection device 10 are the same as those of the first embodiment, the description will be omitted here.
[Image Synthesis Process]
In order to perform processing for deforming one image or processing for deforming and superposing multiple images, processing is required for obtaining colors of pixels from the position of an image before deformation to the position of the deformed image as shown in
Disadvantageously, the processing can be executed by hardware such as a three-dimensional graphics engine, but hardware specialized for DMA access of a rectangular area in a camera module can not process it, for example.
EQ7 is an example of the case that deformation is performed by affine transformation. The (x,y) on the right side indicates the position of the pixel of an image (reference image) after deformation, and the (X,Y) on the left side indicates the position of the pixel of the image (to be superposed) before deformation.
According to this embodiment, an image is divided into small blocks as shown in
The block transfer process by the image synthesis unit 37 will be described below.
In order to calculate the position of a block in an image before deformation, the center position (xc,yc) of the block is calculated (EQ8) from the block starting position (xs,ys) and end position (xe,ye) of the image after deformation, while a coordinate conversion equation as EQ7 is applied (EQ9) to calculate the block starting position (Xs,Ys) and end position (Xe,Ye) of the image after deformation (EQ10).
This method is also applicable to lens distortion correction as in
However, performing the processing by this method deteriorates the quality of the synthesized image. The most problematic deterioration of image quality is a phenomenon that a displacement occurs at the joint of blocks and the discontinuity between the blocks appears which can be seen in
The phenomenon can be eased by changing the position of the boundary of blocks of each image to be superposed as shown in
Some degree of deformation may not require the reduction of the size of the block very much, but, conversely, a high degree of deformation may prevent the processing. In this case, the adaptive change of the size of a block can prevent the deterioration of image quality and allows more efficient processing. The size of all blocks can be changed uniformly as shown in
For example, the equation for determining the block size based on the parameter of affine transformation is as follows:
Here, the value s is a preset value, and a value in the vicinity of one (1), which is a larger than zero (0), is selected therefor. The value s may be set smaller when the processing power is high. However, effect of enhancing the quality of synthesized image can be enjoyed only to a certain small value. That is, if the value s is made smaller than that value, image quality enhancing effect is no more increased. The threshold is larger when blocks can be only moved in units of 2 pixels or 4 pixels due to the processing ability of the hardware. In this case, a large s (such as 2) is more advantageous for the processing speed.
[Correction Failure Determination Process [Off the Rectangle]]
Forming one image by synthesizing multiple images may deteriorate the S/N ratio of the part where the superposition is not performed. In order to prevent this, image clipping must be performed. For maintaining the S/N ratio in a clipping rectangle, the number of the images to be superposed may be same through all the area within the clipping rectangle. When the size and position of the clipping rectangle are specified in advance and if there is an image of which the corrected image does not cover all area of the clipping rectangle, such image is not superposed. This is because it is determined that the correction of such image is failed. The determination method in this case is the same as that of moving picture hand-movement correction according to the third embodiment, which will be described later.
[Correction Failure Determination Process [Rotation]]
A high degree of rotation in superposition in the block synthesis process may deteriorate the clearness of the synthesized image, or adaptive processing may prevent a sufficient speed since the block size is excessively small. In this case, “success” is determined if an optimum block size calculated by EQ12 is higher than a predetermined value min_blocksize while “failure” is determined if lower.
blocksize>min_blocksize (13)
According to this embodiment, a synthesized image is divided into semi-blocks and is transferred by defining the transfer starting position and end position. Thus, even with a displacement component such as a rotation, the effective application of hardware such as a DMA is allowed in memory transfer during image data synthesis, and a clear synthesized image can be generated quickly. In generating a synthesized image by superposing multiple images, the boundaries of blocks can be inconspicuous by dividing transfer blocks at slightly different positions (about 1 to several pixels). Furthermore, the distortion of an image can be corrected by changing the block size depending on degrees of deformation (displacement).
The proportion of synthesis of blocks may be changed in accordance with the intensity and/or the magnitude of a difference in color within blocks. For example, a moving object can be inconspicuous by reducing the proportion of synthesis for a large difference. Conversely, a moving object can be clarified by increasing the proportion of synthesis as the magnitude of the difference increases.
A case that the present invention is applied to a moving picture will be described as the next third embodiment.
[Outline of Hand-Movement Correction in Moving Pictures of High-Degree of Freedom]
In a conventional and general electronic moving picture hand-movement correction (i.e. anti-shake technology), hand-movement correction is achieved by detecting how much a shot frame displaces from a reference frame and outputting a corrected frame by moving the shot frame in parallel based on a detected travel. For example, as shown in
However, in the hand-movement correction technology only with parallel movement, resulting frames are unnatural when a movement in the direction of rotation or scaling occurs as shown in
[Definition of Frame Transforming Data of Moving Picture Hand-Movement Correction with High-Degree of Freedom]
Frame transforming data C from a corrected frame to a shot frame is defined as follows:
Here, (x,y,1)t is the coordinate position of a corrected frame, and (X,Y,1)t is the coordinate position of a shot frame (
where four degrees of freedom (C11, C12, C13, C23) are provided. Notably, the freedom may be called dimension or axis.
[Processing of High Degree of Freedom Moving Picture Hand-Movement Correction]
The processing blocks for moving picture hand-movement correction are as shown in
[Frame Transforming Data Generation Section]
Next, the processing by the frame transforming data generation section 41 will be described.
The processing blocks within the frame transforming data generation section 41 are as shown in
The frame transforming data generation section 41 holds a reference frame Fbase and a reference frame transforming data Cbase as internal data. The reference frame Fbase is a frame on which movement detection is based when new shot data is input. Fbase does not have to hold all frame data (image data) and may hold data required for movement detection only. The reference frame transforming data Cbase is frame transforming data from a corrected frame to Fbase and is held in a 3×3 matrix as well as that of the output frame transforming data C.
Initially, nothing is defined for Fbase (or information indicating the initial state is held therein), and Cbase is set to the standard frame transforming data C0, which is given by:
Here, W and H indicate the width and height of a shot frame, and w and h indicate the width and height of a corrected frame. The standard frame transforming data C0 defined in this way induces transformation to display exactly the center of the shot frame onto the corrected frame (
When the first shot frame is input initially, the frame is set as Fbase first.
When a new shot frame Fin is input to the frame transforming data generation section 41, the motion data detection section 16 first compares it with the reference frame Fbase and outputs motion data M from Fbase to Fin. The motion data M is also a 3×3 matrix like the frame transforming data C. The motion data detection section 16 is a function to be implemented by the units enclosed by the dashed line in
C1=M·Cbase (17)
This C1 expresses frame transforming data from a corrected frame to a shot frame as shown in
The primary frame transforming data C1 is converted to secondary frame transforming data C2 by the frame transforming data standardization section 18. This conversion brings C2 closer to the standard frame transforming data C0 than C1. This is because the state of the standard frame transforming data C0 allows correcting jitter of an image due to a hand-movement in the most stable manner. For example, the frame transforming data as shown in
In this processing, C1 may be strongly brought close to C0 as the difference between C1 and C0 increases. Thus, the speed of the convergence increases, and screens change more naturally when the focus point of a photo-shooting person moves by panning, etc. For example, when
the values of C2 are calculated as follows:
Here, r is a convergence constant and a value equal to or lower than one (1). As the value increases, the speed of convergence to the standard frame transforming data C0 increases.
The secondary frame transforming data C2 is then transmitted to a frame transforming data regularizing section 19 and is converted to the final frame transforming data C. The frame transforming data regularizing section 19 first checks whether C2 is proper frame transforming data or not. The expression, “proper frame transforming data”, refers to one having all of corrected frame coordinates based on C2 included in the shot frame coordinates. For example, the frame transforming data in
More specifically, the points (0,0), (w,0), (0,h) and (w,h) of the four corners of a corrected frame rectangle are required to fit in the shot frame rectangle (0,0) to (W,H) after the transformation based on the frame transforming data C2. In other words, if the condition:
is all satisfied, C2 is determined as being proper. If not, C2 is determined as being not proper. If the data is determined as being proper, the frame transforming data regularizing section 19 directly outputs C2 as output frame transforming data C.
On the other hand, if determined as being not proper, the minimum s (which is a value equal to or higher than 0) satisfying the equation below:
is calculated, and the frame transforming data C is output by using the value of s.
Finally, the reference frame/frame transforming data update section 20 updates the reference frame Fbase and reference frame transforming data Cbase from the shot frame Fin and frame transforming data C such that:
Cbase=C (22); and
Fbase=F (22)
Here, the update is performed when one or a combination of:
Condition 1: The difference between Fbase and Fin is equal to or larger than a threshold;
Condition 2: The difference between C and C0 is equal to or larger than a threshold; and/or Condition 3: At certain intervals (such as every 15 shot frame)
is satisfied. The reason why Fbase and Cbase are not switched for each frame is it prevents slight movement or vibrations of a corrected frame even when no hand-movement occurs due to an error in movement detection when C is a value close to C0 and when update is performed for each frame.
Condition 1 is for switching scenes and is provided because movement detection does not operate normally when a reference frame is not updated upon scene switching.
Condition 2 is provided since movement detection is difficult when the difference between C and C0 is large.
According to this embodiment, since frame transforming data is generated by using the motion data calculation unit 35 according to the first embodiment, a highly accurate corrected frame is generated in real time.
Since the switching of reference frames must be performed fast particularly in handling moving pictures, a reference image defining unit 39 may be provided in the motion data detection section 16 as shown in
The present invention is not limited to the embodiments, and various changes may be made thereto. It is apparent that the methods and steps according to the embodiments can be protected as method for calculating motion data and/or programs for causing a computer to execute a motion data calculation process.
Number | Date | Country | Kind |
---|---|---|---|
2006-048581 | Feb 2006 | JP | national |