Patent Application
20150350483
The present disclosure relates generally to digital imaging and, more particularly, to compressing digital images having varying levels of image noise.
Digital imaging devices appear in handheld devices, computers, digital cameras, and a variety of other electronic devices. Once a digital imaging device acquires an image, an image processing pipeline may apply a number of image processing operations to generate a full color, processed, compressed image in a standardized image format.
While advances in imaging technology enable ever increasing image quality, storage and data transmission concerns, especially for portable devices, continue to drive image and video compression standards to achieve smaller file sizes with increased fidelity. Strategies to achieve smaller file sizes quickly and with minimal loss in fidelity are therefore desirable.
Methods, devices, and systems for selecting an image compression ratio are disclosed. Compression of a digital image can reduce image quality due to reduction in data and detail and the introduction of compression artifacts. While compression may significantly impact the visual quality of an image having high signal-to-noise ratio (SNR), the visual quality of an uncompressed image that already includes a certain degree of image noise due to a low SNR may not be significantly impacted by compression. As such, when storing digital images, images that are identified as having relatively higher degrees of image noise/low SNR may be compressed to a greater degree than images having a relatively lower degree of image noise/high SNR without significant losses in visual quality.
Image analysis needed to measure the amount of image noise in an image can be calculation intensive and time consuming. However, the amount of image noise in a digital image may be correlated with other image characteristics, for example, gain and lux, which may be easily obtained from an image processing pipeline data flow in modern digital image capture devices (e.g., digital camera). Often such characteristics and values are included in the metadata associated with an image. As such, in one embodiment, a compression metric indicative of the degree of image noise in an image is determined based on image characteristics that correlate with image noise. The compression metric may be used to select a compression ratio, such that images having higher image noise/lower SNR may be compressed to a greater degree than images having lower image noise/higher SNR.
In another embodiment, an image may be segmented into a number of regions. A separate compression metric and corresponding compression ratio may be determined for each region. In this manner, higher-image noise/low SNR regions of an image may be compressed to a greater degree, while the higher visual quality of lower-image noise/high SNR regions of the image may be preserved by use of a lower compression ratio for those regions.
Methods, devices, and systems for selecting an image compression ratio are disclosed. A compressed high quality digital image has lower visual quality as compared to the original digital image. However, compression of a low-quality, noisy digital image may not greatly affect the overall image quality. As such, when storing digital images, the image file size may be minimized (or substantially reduced) by selecting a compression ratio based on the amount of image noise in the image, where the compression ratio reduces the image size without significantly impacting the visual quality of the image. Full noise analysis of an image can be calculation-intensive and time consuming. However, the image noise in a digital image may be correlated with other image characteristics (e.g. gain and lux), which are often values included in the metadata associated with an image or otherwise easily determined along a camera's image pipeline. By estimating the amount of image noise in the original digital image based on these characteristics a compression ratio can be selected that reduces the image size but does not significantly impact the image's visual quality.
In one aspect, the compression ratio is based on a compression metric that may itself be based on one or more image characteristics correlated with image noise. The image characteristics may be readily available from an image processing pipeline, reducing the computation required to estimate image noise as compared to traditional noise analysis methods. In an embodiment, the gain or lux may be used as the basis of the compression metric informing the selection of a compression ratio. For compression metric values indicating a high degree of image noise, a higher compression ratio may be used, producing a lower-fidelity compressed image. For compression metric values indicating a low degree of image noise in an image, a lower compression ratio may be used, producing a higher-fidelity compressed image. The relationship between the compression metric and compression level may be continuous or discrete.
In another aspect, different compression ratios may be applied to different regions of an image. In an embodiment, an image is segmented into a number of regions, with the described compression metric analysis applied to each of the regions independently. This allows compression of noisy regions of the image having low local SNR to a larger degree than less-noisy regions of the image, which may enable further reduction in the image file size without significant losses in visual quality.
In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the inventive concept. As part of this description, some of this disclosure's drawings represent structures and devices in block diagram form in order to avoid obscuring the disclosed embodiments. In the interest of clarity, not all features of an actual implementation are described in this specification. Moreover, the language used in this disclosure has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter; rather, the claim language determines such inventive subject matter. Reference in this disclosure to “one embodiment” or to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the disclosed subject matter, and multiple references to “one embodiment” or “an embodiment” should not be understood as necessarily all referring to the same embodiment.
It will be appreciated that, in the development of any actual implementation (as in any development project), numerous decisions must be made to achieve the developers' specific goals (e.g., compliance with system- and business-related constraints), and that these goals may vary from one implementation to another. It will also be appreciated that such development efforts might be complex and time-consuming, but would nevertheless be a routine undertaking for those of ordinary skill in the design of an implementation of image processing systems having the benefit of this disclosure.
By way of example, the image sensor 110 may include a CMOS image sensor (e.g., a CMOS active-pixel sensor (APS)) or a CCD (charge-coupled device) sensor. Generally, the image sensor includes an integrated circuit having an array of pixels, wherein each pixel includes a photodetector for sensing light. As those skilled in the art will appreciate, the photodetectors in the imaging pixels generally detect the intensity of light captured via the camera lenses. However, photodetectors, by themselves, are generally unable to detect the wavelength of the captured light and, thus, are unable to determine color information.
Accordingly, the image sensor may further include a color filter array (CFA) that may overlay or be disposed over the pixel array of the image sensor to capture color information. The color filter array may include an array of small color filters, each of which may overlap a respective pixel of the image sensor and filter the captured light by wavelength. Thus, when used in conjunction, the color filter array and the photodetectors may provide both wavelength and intensity information with regard to light captured through the camera, which may be representative of a captured image.
In one embodiment, the color filter array may include a Bayer color filter array, which provides a filter pattern that is 50% green elements, 25% red elements, and 25% blue elements. For instance, a 2×2 pixel block of a Bayer CFA includes 2 green elements (Gr and Gb), 1 red element (R), and 1 blue element (B). Thus, an image sensor that utilizes a Bayer color filter array may provide information regarding the intensity of the light received by the image capture device at the green, red, and blue wavelengths, whereby each image pixel records only one of the three colors (RGB). This information may be referred to as “raw image data” or data in the “raw domain,” illustrated as block 120 in
Raw image data 120 may then be processed by image signal processor (ISP) 130 to produce a digital image file 140. Image signal processor 130 may perform a variety of image processing functions, such as, for example, a Bayer transformation, demosaicing, noise reduction, white balancing, application of digital gain, and image sharpening. For example, image signal processor 130 may use one or more demosaicing techniques to convert the raw image data 120 into a full color image, generally by interpolating a set of red, green, and blue values for each pixel. Image signal processor 130 may also compute statistical properties of the image that can be used to measure the amount of image noise, including local SNR estimates.
In one embodiment, the image signal processor 130 determines a compression metric associated with an image. In another embodiment, the compression metric is determined by a processor separate from the image signal processor 130, i.e., not by image signal processor 130. The compression metric may be correlated with the amount of image noise in an image. Noise is unwanted signal in an image, or signal that is not indicative of what is being measured. While noise always exists, it becomes visually evident in an image where the signal-to-noise ratio (SNR) is low. The SNR is low where the difference is small between an image sensor signal indicative of the imaged scene and a random signal, sensor noise, or light fluctuations also experienced by the image sensor. Visual image noise often originates, for example, in low-light situations, where the low amount of light hitting the sensor is on the same order as the noise signal generated by the sensor. As such, references to “image noise” herein refer to situations where the SNR for an image is low, such that the noise signal of some magnitude is evident in the image data.
In one embodiment, the compression metric informs the selection of a compression ratio used for the compression 150 of the digital image 140. The determination of the compression metric will be described in greater detail below, with respect to
In one embodiment, digital image file 140 includes image data and metadata. Digital image file 140 may be in a variety of uncompressed or losslessly compressed formats, for example, RAW and PNG. In one embodiment, the metadata associated with digital image file 140 includes the image characteristics discussed above with respect to the raw image data 120. In another embodiment, the metadata associated with digital image file 140 includes a compression metric correlated with image noise.
In block 150, the output of image signal processor 130 is compressed. That is, compression 150 reduces the overall file size of the digital image 140. For example, reducing the number of bytes required to store the digital image. Compression 150 may be lossless, whereby full fidelity of the image is preserved despite reduction in file size, or lossy, where image information is lost, that is, fidelity is reduced. Lossy compression 150 encompasses any method that reduces the size of a digital image file 140. Such methods may employ chroma subsampling, transform coding, predictive coding, quantization, entropy coding, and the like. Lossy compression 150 may, in particular, occur by a variety of industry standard methods, including JPEG, JPEG2000, JPEG-XR, and HEVC still image profile.
In one embodiment, compression 150 occurs according to a selected compression ratio. In one embodiment, the compression ratio may be selected by the image signal processor 130, and sent as an instruction to the compression process 150. In another embodiment, the compression ratio may be selected during the compression process 150 based on a compression metric correlated with image noise. In yet another embodiment, the compression ratio may be selected based on a compression metric by a separate processor and then communicated to the compression process 150.
While pipeline 100 is described with respect to a single digital image, one of ordinary skill in the art will recognize that it is equally applicable to video sequences. For example, a variety of characteristics may be associated with a video file. A compression metric may be derived from video characteristics that are correlated with image noise in the video image or sound. In an embodiment, a video compression ratio may be selected based on the determined compression metric. Video may be compressed according to a variety of industry standard methods, for example, MPEG, H.264, and HEVC.
In block 320, a compression metric indicative of the degree of image noise in the image is determined, according to one embodiment. In one embodiment, the compression metric is derived from at least one image characteristic associated with the image. A variety of image characteristics may be correlated with the amount of image noise in an image. For example, images shot in low light conditions typically have a low SNR, which can result in high image noise. When little light is available for the sensor to generate an image signal, the image gain is typically increased. Where the gain is very high, not only is that part of the captured signal indicative of the image amplified, due to the low SNR, the sensor's noise is also amplified. As such, a high gain typically indicates the presence of image noise in an image.
The image characteristic lux may also be used to identify images that are likely to have image noise. Lux is the amount of light that hits a surface per unit area. As such, a low lux value, such as in low-light situations, has many of the same indications as a high gain value. That is, if an image's lux is low, the image sensor is exposed to a low amount of light. And again, where the signal is small, the image noise and signal are more likely to be confused. Therefore, a high gain value or a low lux value may be indicative of an image that is likely to have a lot of image noise. The gain and lux values for an image are typically stored as metadata, and thus more easily accessible as a noise metric than information obtained by analyzing the image data to identify actual image noise.
Therefore, a compression metric based on an image characteristic such as gain may be indicative of the amount of image noise in an image. In one embodiment, an individual image characteristic may serve directly as the compression metric. For example, the value of the digital gain of the image may be used as the compression metric value. In another example, the lux value for an image may be used as the compression metric value. In one embodiment, a single image characteristic may be mapped onto a compression metric scale, or normalized to determine the compression metric value. For example, the gain value may be divided by the maximum gain value for the camera system, in order to obtain a unit-less compression metric value.
More than one image characteristic may also be used to calculate the compression metric. In one embodiment, each image characteristic used for the compression metric calculation is indicative of image noise in an image. A compression metric function may be developed such that a number of image characteristics contribute to the compression metric value. In one embodiment, each image characteristic value has a coefficient that adjusts the weight and/or units of the characteristic value so that each image characteristic contributes in a meaningful way to the compression metric value. For example, if gain is highly indicative of image noise, while white-point is only slightly indicative of image noise, the gain coefficient may be selected so that the gain component has a greater influence on the compression metric value. Some image characteristics, such as lux, are inversely related to image noise—that is, the greater the lux, the lower the image noise is likely to be. As such, it may be necessary to adjust the compression metric in order to obtain the correct relationship between the compression metric and the compression ratio. An exemplary compression metric function is shown as Equation 1.
metric value=ƒ(gain, lux, . . . )=ωGG+ωluxL+ . . . , (1)
where ωG represents a weighting factor for the gain, G represents the gain, ωL represents a weighting factor for the lux, L represents the lux.
In one embodiment, the compression metric function may be experimentally derived, for example, by performing a regression analysis of various image characteristics. As with single image characteristic values, the value generated by a compression metric function may be normalized, according to one embodiment. For example, the compression metric function may be designed to generate a compression metric value on a scale from 0 to 100.
Returning to
The relationship between the selected compression value and compression metric for an image may be defined so that the high visual quality of low image noise/high SNR images is preserved by using a compression ratio that is low on the full scale of compression ratios, while high image noise/low SNR images having low visual quality are compressed at a ratio that is high on the compression scale, as the loss of data due to compression may not significantly impact the visual quality of the noisy image as compared to the uncompressed version of the image.
In Plot 425, a number of compression metric thresholds 430A-F define the relationship between the compression metric and the selected compression ratio, according to one embodiment. As the compression metric value increases, indicating an increasing degree of image noise in the image, the compression ratio may also increase. By using a number of compression metric thresholds, the degree of compression may be tailored to the degree of image noise in an image. Compression metric thresholds 430A-F are illustrated at regular intervals, with the compression ratio increasing the same amount at each compression metric threshold. However, the compression metric thresholds may be defined at any interval, and the corresponding compression ratio may increase by any amount. The compression metric thresholds may be informed, for example, by a perceptual model, which describes how a viewer's perception of an image is affected by various degrees of image noise.
In Plot 435, the relationship between the compression metric and the compression ratio is continuous, according to one embodiment. Though the relationship is illustrated as linear, it may have any shape whereby the compression ratio generally increases with an increasing compression metric. In addition, the relationship between the compression metric and the compression ratio may be a combination of continuous and discrete, i.e. involving compression metric thresholds.
Returning to
In another embodiment, an image may be divided into regions of varying areas. The area and location of the regions may be based on a number of factors, such as the identification of an image subject, variances in image characteristics (e.g. gain or lux), or otherwise identified as areas likely to have high image noise. Image 730 includes regions of different areas, according to one embodiment. Image 730 includes larger background regions 740 and smaller subject regions 750, according to one embodiment. Subject regions 750 include the subject 710 of the image. In this example, the background of the image may have less detail, less focus, and/or greater image noise as compared to the subject 710. As such, the background regions 740 may have a higher compression metric than subject regions 710, and therefore may be compressed at higher compression ratios than those at which subject regions 710 are compressed. It is to be understood that a variety of methods can be used to define any number and/or size of regions within an image such that the different regions are each compressed using a compression ratio based on a compression metric particular to an individual region.
Returning to
By way of example, the electronic device 800 may represent a block diagram of a portable electronic device, such as a mobile phone or tablet computer system, or similar electronic devices, such as a desktop or notebook computer systems with similar imaging capabilities. It should be noted that the main image processor 835 block, the processor(s) 805, and/or other data processing circuitry generally may be referred to as data processing circuitry. Such data processing circuitry may be embodied wholly or in part as software, firmware, hardware, or any combination thereof. Furthermore, the data processing circuitry may be a single contained processing module or may be incorporated wholly or partially within any of the other elements within electronic device 800. Additionally or alternatively, the data processing circuitry may be partially embodied within electronic device 800 and partially embodied within another electronic device connected to device 800.
In the electronic device 800 of
The image capture element 825 may capture frames of raw image data of a scene, typically based on ambient light. When ambient light alone is insufficient, the strobe 830 (e.g., one or more light emitting diodes (LED) or xenon strobe flash device) may temporarily illuminate the scene while the image capture element 825 captures a frame of raw image data. In either case, the frame of raw image data from the image capture element 825 may be processed before being stored in the memory 810 or non-transitory storage 815 or displayed on the display 820.
In particular, the illustrated image capture element 825 may be provided as a digital camera configured to acquire both still images and moving images (e.g., video). Such an image capture element 825 may include a lens and one or more image sensors configured to capture and converting light into electrical signals and is converted into a raw Bayer, RGB, or YCbCr format, as discussed above. Frames of such image data from the image capture element 825 may enter the main image processing 835 for processing. In some embodiments, the image signal processor 835 may include a dedicated hardware image signal processor and may include one or more dedicated graphics processing units.
The raw image data from the image capture element 825 also may be stored in a framebuffer in the memory 810 accessible to an alternative image processing capability of the electronic device 800. Alternative image processing denotes image processing performed apart from the main image processor 835, and includes image processing performed instead of, or in addition to, processing at the main image processor 835. Consequently, the term also includes processing performed outside of, but in support of, processing of image data by the main image processor 835.
Such an alternative image processing capability of the electronic device 800 may include, for example, image processing or image analysis running in software on the processor(s) 805. Additionally or alternatively, the alternative image processing capability of the electronic device 800 may include other hardware or firmware capable of analyzing the raw image data for certain characteristics. Furthermore, alternative image processing may contribute to the determination of a compression metric, selection of a compression ratio, or compression of an image, as disclosed herein.
The I/O interface 840 may enable electronic device 800 to interface with various other electronic devices, as may the network interfaces 845. These network interfaces 845 may include, for example, interfaces for a personal area network (PAN), such as a Bluetooth® wireless communication network, interfaces for a local area network (LAN), such as an 802.11x Wi-Fi network, and/or interfaces for a wide area network (WAN), such as a 3G or 4G cellular network. Through the network interfaces 845, the electronic device 800 may interface with other devices that may include a strobe 830. The input structures 850 of the electronic device 800 may enable a user to interact with the electronic device 800 (e.g., pressing a physical or virtual button to initiate an image capture sequence). The power source 855 of the electronic device 800 may be any suitable source of power, such as a rechargeable lithium polymer (Li-poly) battery and/or an alternating current (AC) power converter.
It is to be understood that the above description is intended to be illustrative, and not restrictive. The material has been presented to enable any person skilled in the art to make and use the invention as claimed and is provided in the context of particular embodiments, variations of which will be readily apparent to those skilled in the art (e.g., some of the disclosed embodiments may be used in combination with each other). In addition, it will be understood that some of the operations identified herein may be performed in different orders. The scope of the invention therefore should be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled. In the appended claims, the terms “including” and “in which” are used as the plain-English equivalents of the respective terms “comprising” and “wherein.”
| Number | Date | Country | |
|---|---|---|---|
| 62006002 | May 2014 | US |
