`
`____________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`____________
`
`Google Inc.
`
`
`
`Petitioners,
`
`v.
`
`Vedanti Systems Limited
`
`Patent Owner.
`
`____________
`
`Case No. IPR2016-00215
`
`Patent No. 7,974,339
`
`____________
`
`
`
`____________
`
`DECLARATION OF Dr. Omid Kia
`
`
`
`
`
`
`
`Dr. Omid Kia Declaration
`
`I, Omid Kia, make this declaration in connection with the proceedings
`
`identified above.
`
`Introduction
`
`1.
`
`I have been retained by counsel for Vedanti Systems Limited
`
`(“Vedanti”) as a technical expert in connection with the proceedings identified
`
`above. I submit this declaration on behalf of Arendi in the Inter Partes Reviews of
`
`United States Patent No. 7,974,339 (“the ‘339 patent”) in Consolidated Case No.
`
`IPR2016-00212.
`
`2.
`
`I base my opinions below on my professional training and experience
`
`and my review of documents and materials produced in these Inter Partes reviews.
`
`My compensation for this assignment is $450 per hour. My compensation is not
`
`dependent on the substance of my opinions or my testimony or the outcome of the
`
`above-identified proceedings.
`
`Qualifications
`
`3. My qualifications for forming the opinions set forth in this report are
`
`listed in this section and in Exhibit A attached, which is my curriculum vitae.
`
`Exhibit A also includes a list of my publications.
`
`4.
`
`I am currently the Chief Image Scientist at Northstrat, Inc. (hereinafter
`
`referred to as “Northstrat”). In this capacity, I serve as an expert in all of
`
`1
`
`
`
`Northstrat’s imaging initiative along high technology research and development
`
`and serve as an expert in all of Northstrat’s activities pertaining to remote sensing,
`
`surveillance, and image/signal processing problems.
`
`5.
`
`Prior to joining Northstrat, I served as the Senior Staff Scientist at ITT
`
`Exelis, Space Sciences Division (hereinafter referred to as “ITT Exelis”) in an
`
`exact same capacity. ITT Exelis is a leader in Command, Control,
`
`Communications, Computers, Intelligence, Surveillance, and Reconnaissance
`
`(C4ISR) related products and systems and information and technical services,
`
`supplying military, government, and commercial customers in the United States
`
`and globally. ITT Exelis is a pure-play aerospace, defense, and information
`
`solutions company with strong positions in enduring and emerging global markets,
`
`some 20,000 employees, and 2010 revenues of about $6 billion.
`
`6.
`
`In early 2015 ITT Exelis merged with Harris corporation with
`
`complementary capabilities and for further reading please see:
`
`https://www.harris.com/solutions, https://www.harris.com/what-we-
`
`do/intelligence-surveillance-and-reconnaissance, https://www.harris.com/solution-
`
`grouping/remote-sensing-systems and https://www.harris.com/solution/advanced-
`
`analytics.
`
`7.
`
`A copy of my C.V. is Ex. 2002.
`
`2
`
`
`
`8.
`
`In brief, and with particular relevance to the subject matter of this
`
`Inter Parties Review, my background and qualifications to be an expert witness in
`
`this matter are as follows. I received my Ph.D. degree in Electrical Engineering
`
`from the University of Maryland at College Park in 1997. My thesis addressed the
`
`compression and processing of images and has content spanning image processing,
`
`compression and communication theory and application.
`
`9.
`
`Immediately after my graduation, I continued work in media
`
`compression and processing at the Compression Group in the Information
`
`Technology Laboratory of the National Institutes of Standards and Technology. In
`
`this role I continued my research and expanded on similar topics across several
`
`media forms. In particular I served as the United States Government ambassador
`
`to the MPEG standardization group. I also expanded on my thesis research topic to
`
`exploit compressed-domain processing of media for images, video and multimedia
`
`content. I worked with engineers who utilized the MPEG standardization body’s
`
`source code for performance, quality and testing. There were at least three
`
`individuals in my immediate compression group and about 10 individuals working
`
`in other companies that contributed to the source code. These individuals were
`
`degreed with a Bachelor’s degree in either Electrical Engineering or Computer
`
`Science and had direct experience with image processing, compression and
`
`communications as it pertains to encoding and decoding of images. At least two
`
`3
`
`
`
`individuals in my immediate group were recent graduates who were performing a
`
`cross-country internship from France who were interested in pursuing advanced
`
`degrees and part of their education process had to perform actual work to gain
`
`relevant experience. These individuals had a bachelor’s degree in either Electrical
`
`Engineering or Computer Science with one to two years working experience on
`
`compression algorithms that was being studied in my group that included image
`
`compression and transmission.
`
`10. Since 1999, I have held technical leadership positions in image and
`
`signal processing fields for delivery of highly technical solutions to the market. In
`
`particular, I have worked with X-ray imaging for medical diagnostic, security
`
`scanning, and non-destructive testing purposes.
`
`11. From 1999 to 2001, I served as the Chief Technical Officer at
`
`IMACOM, a medical imaging company that manufactured and sold Fluoroscopy
`
`and Radiography systems. From 2002 to 2003 I served as the president at Sigma
`
`Vision, a company that specialized in radiography for veterinary, security, and
`
`non-destructive testing. I worked with hardware and software engineers that
`
`provided technical support in development and maintaining of the imaging
`
`machines produced by the company. A large part of the system was to provide
`
`real-time processing, display and transmission of images within the system and
`
`across system boundaries to systems such as Radiological Information Systems
`
`4
`
`
`
`(RIS) and Hospital Information Systems (HIS). Specifically interface to RIS
`
`systems were accomplished by the Picture Archiving and Communication System
`
`(PACS). We performed research and development in various compression and
`
`transmission options that included a well accepted practice of transmitting Motion
`
`JPEG video where series of frames in a video is compressed by a JPEG image
`
`compression technique. The engineers working with me had bachelor’s degree in
`
`Electrical Engineering and Computer Science with hands on experience in various
`
`aspects of the system some of which was at least one to two year equivalent
`
`experience in image processing, compression and communications.
`
`12. From 2004 to 2009, I served as the Chief Scientist and Director of
`
`Digital X-Ray Development at Imaging Sciences International, Inc. (hereinafter
`
`referred to as “ISII”), a company that marketed a broad scope of imaging and
`
`workflow solutions for the dental market. Since 2009 I have served at ITT Exelis
`
`as a senior scientist responsible for government contracts as Chief Engineer and
`
`Chief Scientist. Also at ITT Exelis, served as one of the leading experts in image
`
`and video compression.
`
`13.
`
`In these capacities, I have worked on many aspects of digital imaging
`
`for X-Rays, optical, hyperspectral, radiofrequency in active or passive modes,
`
`including algorithm development, software development, hardware development,
`
`receptor development, and system design. I have also implemented image and
`
`5
`
`
`
`video compression baselines for various products. In every case where
`
`compression was involved, I had to perform an image quality analysis to determine
`
`the appropriate level of compression or mitigation of the anticipated induced loss
`
`in the compression. Moreover, I had to also perform rate analysis to determine the
`
`level of appropriate compression to achieve in order to meet the available channel
`
`requirements ranging from dial-up modem to satellite communication.
`
`Information Considered in Forming Opinion
`
`14.
`
`I have reviewed the Inter Partes Review pleadings for the above
`
`referenced proceedings including ‘‘339 Patent, U.S. Patent No. 4,791,486
`
`(“Spriggs”) U.S. Patent No. 5,225,904 (“Golin”), U.S. Patent No. 6,529,634
`
`(“Thyagarajan”), “Spatially Adaptive Subsampling of Image Sequences”
`
`(“Belfor”) and Dr. Grindon’s Declarations. I have relied on my own knowledge
`
`and experience as well as published documents in forming my opinions.
`
`Person of Ordinary Skill in the Art
`
`15.
`
`In my opinion, a person of ordinary skill in the art pertaining to the
`
`'339 patent at the relevant date discussed below would have at least a technical
`
`degree in Electrical Engineering, Computer Science or equivalent curriculum with
`
`coursework in image processing and at least one year of hands on experience with
`
`compression and communication techniques.
`
`6
`
`
`
`16. Alternatively, the person of ordinary skill may have earned a degree in
`
`Electrical Engineering, Computer Science or equivalent curriculum with
`
`coursework in compression and communication and at least one year of hands on
`
`experience in imaging.
`
`17. Despite the focus of the invention on reducing bandwidth
`
`requirements, i.e, data reduction, Petitioner’s statement of one of ordinary skill is
`
`so expansive it encompasses persons with no coursework and no experience in data
`
`compression.
`
`‘339 Patent
`18. Coincident with the explosive growth of the Internet through the
`
`utilization of embedded multimedia in the World Wide Web, the invention of the
`
`‘339 patent addressed a demanding and important telecommunication requirement
`
`for image and video streaming. The concept of motion estimation had already
`
`formed the basis for video encoding by achieving high compression rates by
`
`simply removing redundancies associated with moving objects in the scene. With
`
`increased memory, processing and communication improvements a new era of
`
`streaming media was ushered in with video streaming being the most valuable
`
`entity.
`
`19. To address one aspect of managing the transmission requirements for
`
`high throughput and low latency, the inventors of the ’339 patent came up with a
`
`7
`
`
`
`solution of applying a pixel selection process to regions generated by an analysis
`
`process performed on the pixels of a video image.
`
`20. By separately using both of these processes, the resulting system
`
`allows for optimization of the regions and, in addition, a pixel selection process
`
`that “allows random, sequenced, or other suitable processes to be used to select and
`
`locate pixels with optimized regions.” ’339 patent, Ex. 1001, 7:7-9. Combining
`
`frame analysis to generate regions and pixel selection for each region as taught in
`
`the patent has been used to significantly improve the quality of video transmission
`
`and reception over the Internet.
`
`Terminology
`21. The term “data” is used in the ’339 patent in the computing and
`
`communications sense of the word.
`
`22. Thus, as would be understood by one of ordinary skill in the art, data
`
`refers to “digital information” or “bits that can be made available for storage,
`
`transmission and/or interpretation.”
`
`Spriggs
`23. The patent of Spriggs discloses a method for image transmission
`
`employing a process for determining when to subdivide picture areas for reducing
`
`the amount of data needed for transmitting an image. Ex. 1005 at Abstract
`
`(“Spriggs”). Spriggs begins with an image and then determines whether to
`
`8
`
`
`
`subdivide the image into smaller picture areas based on how well an estimated
`
`version of the data compares to the actual pixel data for a given area.
`
`24.
`
`If the estimated version is within a threshold of the actual pixel data,
`
`then the image is not subdivided but if the estimated version is above a threshold
`
`when compared to the actual values, the image is divided both horizontally and
`
`vertically into quadrants.
`
`25. The estimated version is determined using the four corner pixel values
`
`for each area. Spriggs interpolates between the four corner pixel values to create
`
`an estimate for each addressable pixel within the area. The process continues until
`
`all areas of the image that are not subdivided are analyzed to see if they need to be
`
`subdivided.
`
`26. Spriggs produces an encoded data stream that can be decoded by a
`
`receiver by transmitting division codes and also transmitting the sample values
`
`used in the estimation/interpolation process for making the subdivision decisions.
`
`27. Fig. 4 of Spriggs shows how the variably-sized areas are created in
`
`more detail. Spriggs begins by generating the addresses of the four corner points
`
`for an image. Ex. 1005 at Fig. 4. Next, the corner addresses are pushed onto a
`
`memory stack. The corresponding corner samples are then transmitted to the
`
`receiver.
`
`9
`
`
`
`
`
`28. Spriggs then performs an algorithm to determine whether to subdivide
`
`the image. Spriggs performs this methodology by pulling the four corner addresses
`
`off the stack and then generates interpolated samples for the frame based upon the
`
`four corner samples that correspond to the four corner addresses. Ex. 1005, 2:27-
`
`35. The interpolated version of the image is compared with the actual samples to
`
`determine if the estimated and actual samples differ by less than a threshold
`
`criterion. Id., 2:36-46. If the threshold is met then a division code of “0” is
`
`transmitted. Id., 3: 2-4. The “0” indicates that the image has not been subdivided
`
`and that a region has been generated.
`
`10
`
`
`
`29.
`
`If the comparison is not within the margin of the threshold criteria, the
`
`frame is subdivided into quadrants. Ex. 1005, 3:4-7. Five addresses, the address at
`
`the center of the region being divided and the center of each of the four sides of the
`
`region, are determined as represented below by the bolded values (EFGHI). Id.,
`
`3:7-9.
`
`
`
`30. These five center pixel values will be used as corner values for
`
`evaluating the quadrants. Because the frame is subdivided, a subdivision code of
`
`“1” is transmitted. Ex. 1005, 3:4-5.
`
`31. Then, the sample values for the five addresses are transmitted as
`
`shown and described with respect to Fig. 6. Ex. 1005, 3:63-68.
`
`11
`
`
`
`
`
`32. The top of Fig. 6 of Spriggs shows an exemplary image frame that has
`
`undergone the described encoding process and the bottom of Fig. 6 shows the
`
`exemplary output stream for the image frame, which is transmitted to a receiver.
`
`Id. The output stream consists of sample data (SA, SB, etc.) shown in the middle
`
`column and division codes (i.e. 0,1) shown in the left column.
`
`33. The bracketed data in the right column are for informational purposes
`
`and are not transmitted data. Ex. 1005, 3:65-68. These bracketed corner addresses
`
`are used to show the associated corner addresses that relate to the division codes of
`
`‘0’ and ‘1’.
`
`12
`
`
`
`34. The top of Fig. 6 shows a number of regions including regions defined
`
`by their corner points. The regions defined by corner points (AFEI), (FBIG),
`
`(IGHD) are the upper left, upper right and lower right quadrants of the image.
`
`Additionally, there are several smaller regions defined by corner points (EPOS)
`
`(PKSQ) (OSJR) (SQRN) (KINL) (JNCM) AND (NLMH).
`
`35. The method for creating the data stream of Fig. 6 begins as described
`
`above for Fig. 4 with the four corner sample values for the image (SA,SB,SC,SD)
`
`being transmitted. These values are interpolated to generate an interpolated block
`
`and the block of interpolated values is compared to the actual sample values for the
`
`image.
`
`36.
`
`In the example shown in Fig. 6, the threshold is not met and the image
`
`is subdivided. Because of the subdivision, a “1” is transmitted. Id. The
`
`methodology then determines the five center pixel values “SE, SF, SG, SH and
`
`SI”. These will be available for use as corners for the four sub-regions that are
`
`being created. The specification of previously specified corner points with the
`
`newly defined center points allows specification of corner points for newly divided
`
`sub-regions. The methodology then looks at the top left quadrant using the corner
`
`pixel values SA, SF, SE, and SI from the stack to construct interpolated values for
`
`the quadrant. The methodology compares the interpolated pixel values for the
`
`quadrant to the actual pixel values for the quadrant and in this example, it is
`
`13
`
`
`
`determined that the interpolated pixels meet the threshold criterion and a “0” is
`
`transmitted, indicating that this region (defined by corners AFEI) does not need to
`
`undergo any further subdivisions. Based upon having issued the “region data” of
`
`“0”, no further examination of the pixel data of this region is required or performed
`
`in the coding process.
`
`37. The process continues with the upper right quadrant having pixel
`
`values SF, SB, SI, SG, which are taken off of the stack. These four corner pixel
`
`values are interpolated to define interpolated pixel values for the quadrant, which
`
`are compared to the actual pixel values of the quadrant and again the threshold is
`
`met so a “0” is transmitted. No further examination of the pixel data of newly-
`
`created region FBIG takes place in the coding process.
`
`38. The encoding process continues by evaluating the lower left quadrant
`
`composed of corner pixel values “SE, SI, SC, SH” that are read off of the stack.
`
`The interpolated quadrant values when compared to the actual pixel values for the
`
`quadrant do not meet the threshold and therefore, a “1” is generated for
`
`transmission and the quadrant is further subdivided into smaller sized regions. The
`
`quadrant is divided horizontally and vertically and, as shown in Figs. 4 and 6, the
`
`five center values SJ, SK, SL, SM, SN are transmitted to the receiver. The upper
`
`left sub-quadrant defined by corner points (E,K,J,N) is interpolated by corner pixel
`
`values “SE, SK, SJ, SN” and compared to the actual sample values for that region
`
`14
`
`
`
`and the methodology determines that this quadrant should be subdivided and
`
`therefore, a “1” is generated for transmission. Again, the five center values are
`
`transmitted namely, SO, SP, SQ, SR, and SS.
`
`39. As can be seen in Fig. 6, after the final subdivision of the lower left
`
`quadrant occurs causing transmission of pixel values SO, SP, SQ, SR, and SS with
`
`no additional subdivisions occur for the regions defined by corner addresses
`
`(EPOS), (PKSQ), (OSJR), (SQRN), (KINL), (JNCM), (NLMH), and (IGHD). In
`
`fact, no corresponding pixel values need be transmitted for these regions and Figs.
`
`4 and 6 show only a set of 8 consecutive “0”s being transmitted to the receiver.
`
`Thus, for each of these regions only the subdivision code “0” must be sent. No
`
`further examination of the pixel data of these newly-created regions take place in
`
`the coding process.
`
`Golin
`40. Golin discloses a video compression system. Each region of a video
`
`frame is custom encoded. Ex. 1006, 4:68-5:1. Golin teaches a roughness estimator
`
`for detecting edges in the pixel data and if such edges or “roughness” make the
`
`encoding process unacceptable, the region is split horizontally or vertically. Ex.
`
`20. Petitioner relies on Golin for its disclosure of a roughness estimator. Golin
`
`describes fill methods, such as DPCM, that are quite distinct from pixel selection.
`
`15
`
`
`
`‘339 Patent includes both an Analysis System and a Pixel Selection System in
`Contrast to the Spriggs System that only includes an Analysis System
`41. Embodiments of the ’339 patent disclose a system and method that
`
`uses data optimization for reducing data transmission requirements. Ex. 1001,
`
`3:13-14. The data transmission system includes a frame analysis system and a
`
`pixel selection system. The frame analysis system analyzes the data within the
`
`frames to divide the frame into a plurality of regions defined by region data. Id.,
`
`3:51- 4:11. Not until the region data has been determined for a region and been
`
`received by the pixel selection system can the pixel selection system operate on
`
`that region to select pixel values for that region. The region needs to be known and
`
`received for the pixel selection process to take place. Id. The pixel selection
`
`system receives the region data and uses the region data in selecting one or more
`
`pixels from within the region to transmit to the receiver for reconstruction of the
`
`image. Id., 4:12-14.
`
`42.
`
`In contrast, the Spriggs patent includes only an analysis system for
`
`analyzing whether to split an area using quad-tree decomposition. Spriggs does not
`
`include the required pixel selection system.
`
`43. Spriggs recursively looks at an area of an image to see if the area
`
`needs to be subdivided or if corner pixel values will be sufficient to represent the
`
`area. If the corner pixel values are not sufficient to represent the data in the area
`
`when interpolated, the area is subdivided and five center values (center of the
`
`16
`
`
`
`block being subdivided and centers of each of its four sides) are determined and
`
`transmitted as shown by Spriggs.
`
`44.
`
` Each of these pixel values will be used in the corners of the quadrant
`
`regions being formed. These pixel values are data that is transmitted to the receiver
`
`to be used by a decoder for reconstructing the image and are shared among many
`
`possible regions.
`
`45.
`
` Additionally, a division code (“0” or “1”) is transmitted to the
`
`receiver indicative of whether or not the area is subdivided.
`
`46. A region of the image in Spriggs as shown below in Fig. 6 is only
`
`finalized when a “0” code has been generated. The code is transmitted to the
`
`receiver not to any pixel selection system. Thus, given the lack of any process at
`
`the transmitter that has access to the region data for selecting pixels, Spriggs does
`
`not allow for a pixel selection process based on the region data.
`
`17
`
`
`
`
`
`47. Spriggs fails to teach a pixel selection system that receives region data
`
`and that generates pixel values for each region based on the region data as required
`
`by the independent claims. Spriggs simply performs analysis on the image to
`
`determine whether an image should be subdivided horizontally and vertically and
`
`transmits corresponding division codes. These division codes are never received
`
`by a pixel selection system and the division codes are not used for selecting pixel
`
`values for the region.
`
`48.
`
` The initial pixel values that are transmitted in the output data stream
`
`consist of the four corner values for the image, e.g. SA, SB, SC, SD. Since the
`
`regions have not yet been generated, these values are not the result of any pixel
`
`18
`
`
`
`selection for a region. These values are transmitted before the regions are even
`
`known.
`
`49. The coding process of Spriggs goes on to generate the regions. In the
`
`example, regions AFEI, FBIG, JNCM and IGHD become the regions in which
`
`each of A, B, C and D are located. These regions are not known when the values
`
`have been transmitted for A, B, C and D and in particular JNCM is not known until
`
`after at least two subdivision cycles later. As the coding process continues, sets of
`
`five center pixels may be determined whenever there is a split. As was the case for
`
`the initial pixels, these pixels do not receive an associated region until a “0”
`
`division code is issued.
`
`50. Accordingly, Spriggs fails to teach generating one set of pixel data for
`
`each region based on received region data/optimized matrix data.
`
`51. The lower half of Fig. 6 of Spriggs demonstrates a data transmission
`
`stream that is sent to the receiver for the image shown in the upper half of Fig. 6.
`
`The transmission stream does not present a set of pixel data for any of the regions.
`
`The left column represents the division codes/region data, the middle column
`
`represents the transmitted sample values and the right column, which is not
`
`transmitted are helpful comments indicating the corresponding corners of the block
`
`associated with the individual division codes of 1 or 0.
`
`19
`
`
`
`52. As can be seen in Fig. 6, corresponding pixel data is not generated for
`
`each region based upon the division codes/region data. For example, regions
`
`defined by corners (AFEI) region 1, (FBIG) region 2, (EPOS) region 3, (PKSQ)
`
`region 4, (OSJR) region 5, (SQRN) region 6, (KINL) region 7, (JNCM) region 8,
`
`(NLMH) region 9 and (IGHD) region 10, are indicated by a representative “0” for
`
`the division code shown in the first column. Once the “0” is transmitted, the
`
`coding process moves onto a next region. No pixel selection process for the region
`
`is triggered. Spriggs entirely misses the approach of the ’339 patent, which
`
`combined an analysis to generate region data with a process that receives the
`
`region data and selects pixels on the basis of the region data.
`
`53. The recursive process of Fig. 4 in Spriggs is the entire coding process
`
`wherein the encoding process results in an output stream as shown in Fig. 6 of
`
`Spriggs. No further processing at the encoder occurs after the recursive process of
`
`Fig. 4 has completed. The output produces a single output of data that includes
`
`division codes and the pixel data, as corner pixel values, which were used to
`
`generate the division codes based on a desirable encoding error criteria for the
`
`regions. Neither the division codes nor the pixel data are used in a further “pixel
`
`selection” process. There is no process disclosed or suggested by Spriggs, which
`
`receives the region data and generates pixel data for a region.
`
`20
`
`
`
`54.
`
`In Fig. 4, the corner pixel values can be found among the pixel values
`
`transmitted during the production of the regions by the analysis system. These
`
`values are determined during the region generating process. There is no subsequent
`
`pixel selection process.
`
`55. The recursive flowchart of Fig. 4 determines when to subdivide
`
`regions into sub-regions and provides pixel values for transmission. However, this
`
`is a singular process and the pixel values are transmitted first before a region is
`
`finally identified by a ‘0’ division code.
`
`56. As shown in Fig. 6, SA through SI are transmitted before region AFEI
`
`is determined by the ‘0’ division code. The process of the pixel selection system is
`
`completely missing from Spriggs and the Petitioner merely repurposes its analysis
`
`for the “analysis system” suggesting that it likewise covers the “pixel selection
`
`system.”
`
`57.
`
`In Fig. 4, the “0” and “1” division codes are transmitted to the
`
`receiver. The “1”s and “0”s in the data stream can be used by the receiver to
`
`determine how to reconstruct the image at the receiver. The division codes are not
`
`received and used in a process for selecting pixel data in the transmitter.
`
`58.
`
`In Spriggs, upon each iteration of its recursive process as exemplified
`
`in Fig. 4, the process always begins with the corner pixel values and addresses of
`
`the corner pixels of the initial block. The process performs an analysis function to
`
`21
`
`
`
`determine if the block should be split and be recorded in the division codes. These
`
`division codes are immediately transmitted as they are generated as seen in Fig. 4.
`
`The region data are never received by a pixel selection system in Spriggs encoder
`
`that generates one set of pixel data for each region forming a new set of data for
`
`transmission as required by claim 1. The division codes are transmitted in a
`
`transmission stream to the receiver.
`
`59. The example of corner pixels given in the Petition is “SA, SB, SC,
`
`SD”. Each of these is the corner of a different region in Fig. 6. Spriggs’ coding
`
`process begins with these pixels. They are certainly not selected “based on”
`
`optimized matrix data. They are generated and transmitted before Spriggs’ coding
`
`process makes its first decision.
`
`60. Spriggs never selects between two or more sets of pixel data, which
`
`would require a decision process that is clearly lacking in Spriggs.
`
`61.
`
`In Fig. 6, the set SA, SB, SC, SD does not form a region. SA is in
`
`region AFEI, SB is in a different region FBIG and SC is in yet another region
`
`JNCM.
`
`Belfor
`
`62. Belfor reduces the data representing an image by performing a
`
`subsampling process where “[b]y discarding a part of the pixels, the image can be
`
`transmitted more efficiently.” Ex. 1007, 1. Fixed subsampling is compared with
`
`22
`
`
`
`spatially adaptive subsampling. Improved storage and transmission performance is
`
`shown for spatially adaptive subsampling relative to fixed subsampling.
`
`63.
`
`“In a spatially adaptive subsampling scheme, the image is subdivided
`
`into square blocks, and each block is represented by a specific spatial sampling
`
`lattice.” Ex. 1007, 1. The scheme addresses high detail and low detail areas of an
`
`image by using a dense sampling lattice in high detail blocks and a sparse sampling
`
`lattice with few pixels in a low detail block. Accommodating the varying detail
`
`through variable subsampling in conjunction with a representative lattice structure
`
`provides the improvement over using a fixed sampling and lattice structure.
`
`64. Block sizes considered in Belfor refer to the dimensions of the block.
`
`Belfor illustrates a block size with dimensions of 8 pixels by 8 pixels. Ex. 1007, 4.
`
`Furthermore, Belfor notes, “The number of possible modes is affected by the block
`
`size because for decreasing block size, the number of possible sampling lattices
`
`within the block decreases as well.” Id. at 4. Because the number of pixels in the
`
`block makes it possible to use a certain number of subsampling lattices. For a
`
`given block size, Belfor provides an associated set of modes from which to choose,
`
`each mode having a specific subsampling lattice. Id. at 4. For the use of
`
`sophisticated interpolation in decoding, the set of modes for a given block size is
`
`preferably hierarchical. Id. at 4. Hierarchical means mode n+1 is always a subset
`
`23
`
`
`
`of mode n, as shown in Fig. 4. Id. at 4.
`
`65. The system of Belfor, as shown in Fig. 5, runs each block through a
`
`pre-filtering step. This is followed by a subsampling step for each of the modes in
`
`
`
`the set.
`
`66. The subsampled image for each mode at each block is evaluated for
`
`quality by an error computation module. Then the mode allocation module makes
`
`the determination as to which mode is the best representative mode to be allocated
`
`to each individual block. After the modes have been allocated, in other words – a
`
`
`
`24
`
`
`
`mode has been assigned to each block – the data for the image is transmitted
`
`through the channel. The data would include the mode assignment for each block
`
`and the pixels remaining after applying the respective subsampling to each of the
`
`blocks.
`
`67. Belfor further discusses the complexities involved in allocating the
`
`modes to the blocks. According to Belfor, “The mode allocation is of great
`
`significance as it influences the output quality considerably.” Ex. 1007, 4. In the
`
`simplest allocation approach, only two modes are used. To produce a desired
`
`output bit rate, a fraction of the blocks will be assigned one mode and the
`
`remaining fraction of the blocks will be assigned the other mode. The fractions are
`
`chosen so that the total bits approximate the desired bit rate. To implement this
`
`approach, the less dense mode is applied to all of the blocks. The distortion or
`
`error computation is determined for each block. Then the blocks with the highest
`
`computed distortion will be assigned the denser mode in an effort to increase total
`
`number of bits while decreasing the largest amounts of distortion. It is the aim of
`
`the process to still keep the total number of bits below a desired amount while
`
`decreasing the distortion.
`
`68. Belfor seeks to describe a mode allocation algorithm that can be used
`
`with any number of modes. Ex. 1007, 5. But increasing the modes to more than
`
`two will produce a pair of equations that cannot be uniquely solved. Id. at 5. It is
`
`25
`
`
`
`necessary that the modes produce a rate-distortion curve that is convex as shown in
`
`Fig. 6. Id. at 5. In other words, as the bit rate ratio (remaining bits in the lattice
`
`divided by total bits in the block) increases moving to the right in Fig. 6, the
`
`distortion D is reduced. Id. at 5.
`
`
`
`This convexity is in line with the theoretical definition of rate distortion theory in
`
`that every added bit used to represent an image should meth