Abstract: A deep learning based compression (DLBC) system trains multiple models that, when deployed, generates a compressed binary encoding of an input image that achieves a reconstruction quality and a target compression ratio. The applied models effectively identifies structures of an input image, quantizes the input image to a target bit precision, and compresses the binary code of the input image via adaptive arithmetic coding to a target codelength. During training, the DLBC system reconstructs the input image from the compressed binary encoding and determines the loss in quality from the encoding process. Thus, the models can be continually trained to, when applied to an input image, minimize the loss in reconstruction quality that arises due to the encoding process while also achieving the target compression ratio.

Abstract: Systems and methods for integrated graphics rendering are disclosed. In certain embodiments, the systems and methods utilize a graphics engine, a video encoding engine, and remote client coding engine to render graphics over a network. The systems and methods involve the generation of per-pixel motion vectors, which are converted to per-block motion vectors at the graphics engine. The graphics engine injects these per-block motion vectors into a video encoding engine, such that the video encoding engine may convert those vectors into encoded video data for transmission to the remote client coding engine.

Abstract: Improved transforms are used to encode and decode large video and image blocks. During encoding, a prediction residual block having a large size (e.g., larger than 32×32) is generated. The pixel values of the prediction residual block are transformed to produce transform coefficients. After determining that the transform coefficients exceed a threshold cardinality representative of a maximum transform block size (e.g., 32×32), a number of the transform coefficients are discarded such that a remaining number of transform coefficients does not exceed the threshold cardinality. A transform block is then generated using the remaining number. During decoding, after determining that the transform coefficients exceed the threshold cardinality, a number of new coefficients are added to the transform coefficients such that a total number of transform coefficients exceeds the threshold cardinality. The transform coefficients are then inverse transformed into a prediction residual block having a large size.

Abstract: This application discloses a communications chip structure, including: a channel selection module, configured to receive an input signal, where the input signal is a signal of a preset narrow bandwidth span or a signal of a preset wide bandwidth span; and a digital baseband module, configured to control the channel selection module to select a first sampling and quantification channel when the input signal is a signal of the preset narrow bandwidth span, or control the channel selection module to select a second sampling and quantification channel when the input signal is a signal of the preset wide bandwidth span. The channel selection module is further configured to send the input signal to the first sampling and quantification channel or the second sampling and quantification channel for sampling and quantification.

Abstract: A method of encoding a coding tree unit in a video bitstream. A plurality of candidate configurations are formed for the coding tree unit, each of the candidate configurations having a variation of at least one of a set of partitioning modes and encoding parameters. A candidate configuration is selected from the plurality of candidate configurations based on a predetermined maximum bit rate for the coding tree unit, the selected candidate configuration having a size within the predetermined maximum bit rate. The coding tree unit is encoded using the selected candidate configuration.

Abstract: A video coding or decoding method using inter-image prediction to encode input video data in which each chrominance component has 1/Mth of the horizontal resolution and 1/Nth of the vertical resolution of the luminance component, where M and N are integers equal to 1 or more, including storing one or more images preceding a current image, interpolating a higher resolution version of prediction units of the stored images so that the luminance component has a horizontal resolution P times that of the corresponding portion of the stored image and a vertical resolution Q times that of the corresponding portion of the stored image, detecting inter-image motion between a current image and the one or more interpolated stored images so as to generate motion vectors between a prediction unit of the current image and areas of the one or more preceding images, and generating a motion compensated prediction.

Abstract: A method for graph-based spatiotemporal video segmentation and automatic target object extraction in high-dimensional feature space includes using a processor to automatically analyze an entire volumetric video sequence; using the processor to construct a high-dimensional feature space that includes color, motion, time, and location information so that pixels in the entire volumetric video sequence are reorganized according to their unique and distinguishable feature vectors; using the processor to create a graph model that fuses the appearance, spatial, and temporal information of all pixels of the video sequence in the high-dimensional feature space; and using the processor to group pixels in the graph model that are inherently similar and assign the same labels to them to form semantic spatiotemporal key segments.

Abstract: A system, method, and computer program product are provided for automatic root cause analysis. In operation, a root cause analysis system identifies at least one event associated with one or more records for which to perform a root cause analysis. The root cause analysis system performs a root cause analysis of the event by automatically generating a decision tree based on all records in the current time-window such that each leaf in the decision tree represents the probabilities for class labels of a target variable and each branch in the decision tree represents a feature that leads to a corresponding class label probability. The root cause analysis system automatically generates the decision tree by automatically selecting at each step the feature that maximizes information gain based on a current subset of data.

Abstract: The embodiments of the invention relate to a transmitter apparatus (TA1) for conditioning a multicarrier signal (RFS). The transmitter apparatus (TA1) contains means (FE-PU) for grouping subcarriers of the multicarrier signal (RFS) into a first frequency block, which contains a first group of the subcarriers and into at least a second frequency block, which contains at least a second group of said subcarriers. The transmitter apparatus (TA1) further contains first filtering means (LPF-1) for sideband suppression outside of the first frequency block and at least second filtering means (LPF-2, LPF-M) for simultaneous and separate sideband suppression outside of the at least second frequency block. The embodiments of the invention further relate to a method for conditioning a multicarrier signal (RFS).

Abstract: A system includes a first low noise amplifier, a second low noise amplifier, a local analog oscillator signal, a signal splitter, a mixer, a mixer, an analog to digital converter and a digital channelizer. The first low noise amplifier outputs a first amplified analog signal based on a received analog antenna signal at a time t0. The second low noise amplifier outputs a second amplified analog signal based on the received analog antenna signal at a time t1. The local analog oscillator signal outputs a local analog oscillator signal. The signal splitter outputs a split analog oscillator signal and a split analog oscillator signal. The mixer outputs a first mixed signal. The mixer outputs a second mixed signal. The analog to digital converter outputs a combined digital signal. The digital channelizer outputs a received signal based on the combined digital signal.

Abstract: An apparatus for encoding a video includes: a predictor to predict a current block by using a plurality of prediction unit types and generate one or more predicted blocks; a subtracting unit to generate a residual block by the subtracting the predicted blocks from the current block; a transformer to generate a frequency transform block by transforming the residual block; a quantizer to generate a quantized frequency transform block by quantizing the frequency transform block; and an encoder to encode the quantized frequency transform block into a bitstream. When the quantized frequency transform block for one of the prediction unit types has all zero coefficients, the predictor terminates predicting the current block by using the remainder of the prediction unit types and determines the prediction unit type with said all zero coefficients as the prediction unit type for the current block.

Abstract: A moving image coding apparatus which acquires moving image data represented by one of a plurality of different bit depths, converts the acquired moving image data to moving image data having the lowest bit depth among the plurality of different bit depths, determines a prediction mode relating to intra-picture prediction on the basis of the moving image data of the converted bit depth, and prediction-codes the acquired moving image data represented by one of the plurality of different bit depths in accordance with the determined prediction mode.

Abstract: This disclosure describes frame interpolation techniques within a wavelet transform coding scheme. The frame interpolation may be used to generate one or more interpolated frames between two successive low frequency frames coded according to the wavelet transform coding scheme. Such interpolation may be useful to increase the frame rate of a multimedia sequence that is coded via wavelet transforms. Also, the techniques may be used to interpolate lost frames, e.g., which may be lost during wireless transmission.

Abstract: An encoder is operable to encode input data to generate corresponding encoded output data. The encoder includes data processing hardware. The encoder compresses content associated with blocks or packets, so that the encoded output data is smaller in size than the input data.

Abstract: Using scan conversion processing of changing the scan order for each block, parallel scan conversion processing is executed if possible. A scan status holding unit holds statistical information based on the appearance frequency values of coefficients in a block. A scan order holding unit holds coefficient position information in which the coefficient positions in a block are arranged based on the scan order. A significant data position information generation unit scans one block data in accordance with coefficient position information, and generates information representing nonzero/zero for each data position. Based on the statistical information and the information generated by the significant data position information generation unit, a parallel number determination unit determines whether to process two blocks in parallel or process one block. In accordance with the determination, a scan conversion unit scan-converts two input blocks in parallel or scan-converts only one of the blocks.

Abstract: Provided are an apparatus for multi-stage transforming a plurality of unit blocks in multi-dimension that can improve compression efficiency of video data by collecting Discrete Cosine Transforming (DCT) coefficients of neighboring blocks and performing an additional transformation based on the DCT coefficients of an original picture and a differential picture. The method includes the steps of: performing a Discrete Cosine Transform (DCT) on inputted picture data and selecting R blocks of a predetermined size from DCT picture data, where R is a natural number equal to or greater than 2; arranging DCT coefficients of each of the selected R blocks according to each frequency in one-dimension; and performing one-dimensional transformation again on the DCT coefficients arranged in one-dimension.

Abstract: Included are embodiments for processing video data. At least one embodiment includes receive logic configured to receive the video data chosen from a plurality of formats and filter logic configured to filter the video data according to the instruction. Similarly, some embodiments include transform logic configured to transform the video data according to the instruction, where the instruction contains a mode indication in which the filter logic and the transform logic execute based on the format of the video data.

Abstract: A method of decoding a video signal is disclosed. The present invention includes determining whether to store a first partial picture when the first partial picture and a first full picture are corresponding to a first temporal point and storing the first partial picture for decoding a second full picture referring to the first partial picture, the second full picture being corresponding to a second temporal point, the second temporal point being located after the first temporal point, wherein a level of the first partial picture on a scalable domain is lower than a level of the second full picture on the scalable domain.

Abstract: In decoding a scalable video signal using a partial picture reference on a temporal domain and a scalable domain, the present invention provides a method including obtaining a first partial picture on a first temporal point, and decoding a full picture referring to the first partial picture, the full picture being on a second temporal point, the second temporal point being located after the first temporal point, wherein a level of the first partial picture on a scalable domain is lower than a level of the full picture on the scalable domain.

Abstract: A moving picture coding method includes: determining a luminance change by determining whether it is possible that a luminance change equal to or greater than a predetermined amount has occurred between the reference coded picture and a current picture; determining a flat region by determining whether a degree of non-uniformity in pixel values in a current block is smaller than a predetermined threshold; and determining a quantization width wherein, when it has been determined that it is possible that the luminance change equal to or greater than the predetermined amount has occurred and that the degree of non-uniformity is smaller than the predetermined threshold, a quantization width is determined to be smaller than a quantization width when it has not been determined to be possible that the luminance change equal to or greater than the predetermined amount has occurred or that the non-uniformity degree is smaller than the predetermined threshold.

Abstract: A method for receiving a broadcasting signal and a broadcasting signal receiver are disclosed. Even when a cell is changed while an emergency alert is output, the emergency alert can be continuously output using emergency alert table information included in the broadcasting signal and channel information of the cell. The emergency alert table information may include a cell identifier and the channel information of the cell may include virtual channel information of the cell.

Abstract: A method for receiving a broadcasting signal and a broadcasting signal receiver are disclosed. Even when a cell is changed while an emergency alert is output, the emergency alert can be continuously output using emergency alert table information included in the broadcasting signal and channel information of the cell. The emergency alert table information may include a cell identifier and the channel information of the cell may include virtual channel information of the cell.

Abstract: Embodiments of the invention provide an instruction that computes the horizontal and vertical values (H,V) based upon the predefined equations. Based upon the horizontal and vertical values (H,V) and the current sign bit being processed at [m,n], the output context and decision pair (CX,D) is determined placed into a destination register.

Abstract: Low complexity (16 bit arithmetic) H.264 video compression replaces a single quantization table for all quantization parameters with multiple quantization tables and thereby equalizes quantization shifts and round-off additions; this eliminates the need for 32-bit accesses.

Abstract: A system provides lossless split and merge processes of integer discrete cosine transform (DCT) transformed data such that the discrete cosine transform of one data block may be split into two half length DCT odd and even blocks for merging, with split and merge processes being lossless and are generated in the discrete cosine transformed domain. After splitting, the redundancy existing between the two integer discrete cosine transformed half data blocks allows one to approximately reconstruct the original data block in case one of the discrete cosine transformed half data block is lost during transmission.

Abstract: A method of encoding a set of data representing physical quantities includes the steps of dividing the set of data into subsets, calculating a first encoding cost for each subset using a first encoding mode, calculating a second encoding cost for each subset using a second encoding mode, and selecting an encoding mode per subset as a function of the first and second encoding costs, in which the two encoding costs are calculated according to the same rate-distortion compromise (?), for the image overall.

Abstract: An image processing apparatus for converting image data between a raster format and a block format including an image data processor for providing the image data including a luminance component and at least one chrominance component in the raster format, at least two FIFO memories for storing corresponding image data components, a multiplexer for multiplexing the image data components from the at least two FIFO memories, a line buffer memory for storing outputs of the multiplexer linearly, and an image compressor for receiving the image data components in block format in sequence from the unified line buffer memory and compressing the received image data components.

Abstract: A dynamically scaled file encoding method and apparatus are disclosed. A file encoding system using JPEG encoding can be configured to produce relatively constant compressed file sizes irrespective of the initial file size and file contents. The system retrieves an initial file or image that is to be compressed and determines a target bit rate corresponding to the compressed file. The target bit rate is used to determine an initial scaling factor. The initial file is encoded using a JPEG encoder having coefficients scaled by the initial scaling factor. The resultant bit rate can be adjusted in a second loop if greater than the desired bit rate. To adjust the bit rate, a recomputed scaling factor is determined from the resultant bit rate. The initial file is then encoded with coefficients scaled by the recomputed scaling factor to achieve a bit rate that is within the target bit rate.

Abstract: An image encoding/decoding apparatus is provided, including a frequency band division unit for dividing image data into a plurality of frequency bands to generate sub-band coefficients, a sub-band coefficient interpolation unit for interpolating sub-band coefficients outside a region of a shape by using shape information representing a shape of the image data and the sub-band coefficients in the region of the shape, an encoding unit for encoding the interpolated sub-band coefficients to generate encoded data, and a shape information encoding unit for encoding the shape information to generate shape information encoded data.

Abstract: A method of displaying a thumbnail image of the original image of a data file stored in a storage unit on a display unit includes: (a) storing in the storage unit a first compressed code relating to an image for displaying the thumbnail image of the data file, the first compressed code being generated by dividing the image into a plurality of tiles and performing discrete wavelet transform and hierarchical encoding on the pixel values of the image tile by tile; (b) setting the resolution of the thumbnail image in accordance with the format type of the data file; (c) extracting a second compressed code according to the resolution that was set from the first compressed code stored in the storage unit; and (d) displaying the thumbnail image based on the second compressed code that was extracted.

Abstract: A method for carrying out a step of extracting a portion of a predetermined size from the image data, and a step of preparing a predetermined number of compressed partial data compressed with different compression parameter values by repeating preparation of compressed partial data by compressing the partial data in predetermined format for the predetermined number of times with the compression parameter value used in the predetermined format changed for each compression, on the entire image data. The method also calculates the total of the data sizes of all of the compressed partial data compressed with the same compression parameter value for each of the compression parameter values and prepares the compressed image data file from compressed partial data compressed with a compression parameter value for which the total is not larger than a given threshold and which is the closest value to the threshold.

Abstract: An image processing apparatus is adapted to a surveillance camera system, and a CPU included in the surveillance camera system divides a plurality of continuous screens of an image signal into a plurality of blocks by each screen, and detects a specific object, which is an object with movement, from a luminance change of a Y signal found by a movement detection circuit. In addition, the CPU specifies the block in which a movement of the object is detected, and sets the specified block as an area to be noticed by taking advantage of an ROI function of a JPEG 2000. Next, a JPEG 2000 CODEC is instructed to compress an image of a movement-detected block into a high quality image by an alarm compression rate, and compress the image of a movement-not-detected block by a normal compression rate having a higher compression rate than the alarm compression rate.

Abstract: A method and arrangement for detecting a watermark embedded in an MPEG compressed signal includes a conventional MPEG decoder stripped to such an extent that a modified baseband video signal suitable for watermark detection is obtained. A plurality of pictures with the embedded watermark is accumulated (2,3,4) in the transform domain, and the inverse DCT (5) is applied to the accumulated result. Conventional watermark detection (6) is then applied to the accumulated plurality of pictures in the spatial domain.

Abstract: A method for encoding video with a two-dimensional (2D) transform separable to two one-dimensional (1D) transforms. The method receives an array of values for a sub-section of an image, performs a first 1D-transform of the array, transposes the resulting array, and performs a second 1D-transform of the array resulting from the transpose. The method, without performing another transpose, generates a data stream using a transposed scan order based on the values of the array resulting from the second transform. A method for decoding video encoded by a 2D transform, which separable to two 1D transforms. The method receives a data stream containing encoded values for an image, parses out the values into an array using a transposed scan order, performs a first 1D-inverse transform on the array, transposes the resulting array, and performs a second 1D-inverse transform of the array resulting from the transpose to produce a decoded output.

Abstract: Implementing a two-dimensional inverse discrete cosine transform function includes executing two one-dimensional inverse discrete cosine transforming functions. Each of the one-dimensional functions is controlled to operate on a matrix of coefficients in either of two different directions.

Abstract: A volume data encoder with high encoding efficiency is provided. The object of the coding is volume data which contains a plurality of tomogram planes output from a CT and an MRI. A header analysis unit separates each plane image into header information and pixel information. A header compression unit compresses the separated header information. On the other hand, a two-dimensional transform unit conducts frequency decomposition on the pixel information. A skip portion detection and table generation unit detects skip portions that are the same in all coefficients in a z-direction, and stores them in a table. A one-dimensional transform unit conducts one-dimensional transform on the pixel information for coefficients except the skip portions. A unit block division unit divides each subband into unit blocks. An entropy encoding unit determines a parameter for entropy encoding according to statistical properties of all coefficients in all unit blocks included in each class.

Abstract: A processing device (200) includes three hardware extensions: a motion estimation extension 202, a pixel interpolation extension 204 and a DCT/iDCT extension 206. The hardware extensions perform functions which would otherwise be highly processor intensive, resulting in high power consumption and/or low quality video/imaging processing. The processing device 200 could be used, for example, in a mobile videophone 150.

Abstract: A method and apparatus for compressing image data for storage in a memory device is presented. This is accomplished by separating the image data into a plurality of pixel sets where each pixel set is of a predetermined pixel set size. A discrete cosine transform is then performed on each of the pixel sets to produce a plurality of transform coefficients. These transform coefficients are then compressed to produce a compressed data set. Compressing the transform coefficients preferably includes determining a coefficient set that includes a portion of the transform coefficients that reasonably approximate the pixel set. These coefficients are then mapped to known ranges such that a limited number of bits can encode values throughout these predetermined ranges. The mapped coefficients resulting from the mapping step are then manipulated to fit within a limited number of bits assigned to each coefficient. The limited number of bits is determined partially based on the coefficient set to be compressed.

Abstract: A coding apparatus which can simplify a coding circuit and also simplify the processing of quantization. Image data that consists of R-, G-, and B-color components read from a scanner into buffers is subjected to transformation using a position correlation for each of RGB in position-correlation transformation sections. The DC component of each of the components is then color-converted in a color conversion section to be a DC lightness signal and DC color signals, and the signals are quantized in a quantization section and a coding section. Whereas, the AC components of each of the components are subjected to quantization in a first step in an AC component quantization section, and further to color conversion in a color conversion section to be an AC lightness signal and AC color signals. The signals are quantized in a second step in a quantization section and a coding section.

Abstract: A multi-layer embedded bitstream is generated from a subband decomposition by partitioning each subband of the decomposition into a plurality of blocks; and encoding the blocks of each subband. The blocks of each subband are coded independently of each other. Resulting is a block bitstream corresponding to each block. Truncation points may be identified on the block bitstreams, and selected portions of the block bitstreams may be concatenated, layer-by-layer, to form the single-layer or multi-layer bitstream. Syntax information may also be added to the multi-layer bitstream. An image can be reconstructed from the embedded bitstream at a desired bit-rate or resolution by reading the syntax information, randomly accessing desired portions of the block bitstreams, decoding the randomly accessed portions, dequantizing the decoded portions, and applying an inverse transform to the dequantized portions.

Abstract: A method of reformatting a data stream comprised of frames arranged into superframes for transmission by a block coding scheme is disclosed herein. The payloads of a plurality of frames are concatenated, where a superframe structure is present the frames in the superframe are concatenated, to form a transmission block. The headers of these frames are replaced by a syncword. The present invention can be used in any communications system which uses MPEG frame form, and can be extended any communications system that can benefit from modification to the frame format.

Abstract: Techniques for classifying video frames using statistical models of transform coefficients are disclosed. After optionally being decimated in time and space, image frames are transformed using a discrete cosine transform or Hadamard transform. The methods disclosed model image composition and operate on grayscale images. The resulting transform matrices are reduced using truncation, principal component analysis, or linear discriminant analysis to produce feature vectors. Feature vectors of training images for image classes are used to compute image class statistical models. Once image class statistical models are derived, individual frames are classified by the maximum likelihood resulting from the image class statistical models. Thus, the probabilities that a feature vector derived from a frame would be produced from each of the image class statistical models are computed.

Abstract: A method for decoding video data blocks using variable length codes, comprising transforming information about the spatial frequency distribution of a video data block into pixel values. Prior to said transformation, a first reference value (Xref) representing the abruptness of variations in information about spatial frequency distribution within the block is generated, after said transformation, a second reference value (&Dgr;) representing the abruptness of variation in certain information between the block and at least one previously transformed video data block is generated. The first reference value (Xref) is compared to a first threshold value (TH1) and the second reference value (&Dgr;) to a second threshold value (TH2); and as a response to either of the first (Xref) and second reference values (&Dgr;) being greater than the first (TH1) and respectively the second threshold value (TH2), an error in the block is detected.

Abstract: To obtain real-time responses with interactive multimedia servers, the server provides at least two different audio/visual data streams. A first data stream has fewer bits per frame and provides a video image much more quickly than a second data stream with a higher number of bits and hence higher quality video image. The first data stream becomes available to a client much faster and may be more quickly displayed on demand while the second data stream is sent to improve the quality as soon as the playback buffer can handle it. In one embodiment, an entire video signal is layered, with a base layer providing the first signal and further enhancement layers comprising the second. The base layer may be actual image frames or just the audio portion of a video stream. The first and second streams are gradually combined in a manner such that the playback buffer does not overflow or underflow.

Abstract: The color of a pixel is represented in a pixel storage word, wherein color coordinate data and intensity data are coded separately in two fields of the pixel storage word, the color field and the intensity field. This permits a range of colors to be represented in a relatively small number of bits. In one embodiment of the invention, each of the three color coordinates of an input color is coded separately. The coded color components are concatenated and placed in a coded color field of the pixel storage word. In another embodiment of the invention, the color coordinate data is encoded as a color index, a binary value which maps to color coordinate data according to a predefined color table. Intensity data is also encoded with a binary value. The coded intensity can represent the intensity directly, or can be an index to intensity data stored in a predetermined intensity table.

Abstract: The present invention provides a method of and system for determining whether a local area of a compressed video stream represented by a plurality of DCT encoded blocks subject to decoding and filtering/scaling is a stationary area or an interlaced moving area. Further, given such information, the invention relates to dynamically switching between frame- or field-based operations in a smart way, thus optimizing the output picture quality. Also, a DCT-domain-filtering scheme for field-based filtering/scaling of frame-DCT data is provided herein.

Abstract: Disclosed is a digital broadcasting transmission system which includes: a plurality of single program encoders for encoding and multiplexing respective single programs constituted with video/audio/data signals; a PSIP generator for generating PSIP that provides detailed information on video/audio/data broadcasting programs to be multiplexed in the single program encoders; a multiple program multiplexer for multiplexing digital broadcasting stream transmitted from the signal program encoders and the PSIP transmitted from the PSIP generator; and a transmission unit for transmitting digital broadcasting signals multiplexed in the multiple program multiplexer.

Abstract: A double sampling time-integrating pixel sensor having a photo-detector, a capacitor, a comparator and two pixel data buffers. In operation, the photo-current from the photo-detector charges the capacitor and produces a photo-voltage. The photo-voltage sensed by the capacitor and a first reference voltage are compared by the comparator. When the photo-voltage exceeds the first reference voltage, a first global counter value is latched into the first pixel data buffer. When the photo-voltage exceeds a second reference voltage, a second global counter value is latched into the second pixel data buffer. The optical power falling on the photo-detector is determined from the difference between the second and first counter values. An array of sensors is incorporated into a semiconductor device together with circuitry to read and decode the pixel data buffers.

Abstract: A column address is decoded by each column decoder, and an arrangement of pixels to be written to a memory cell array is rotated by a rotation circuit. The rotation result is written to the memory cell array in accordance with the decoding result of the column address. Then, the column address is decoded by each column decoder to a decoding result which is different from the decoding result of the column address at the time of writing. In accordance with the decoding result, pixels stored in the memory cell array are read. The arrangement of the read pixels is rotated by the rotation circuit and then outputted. Thus, rearrangement of pixels constituting an image can be carried out quickly while avoiding any increase in the size of the device.

Abstract: A linear transform apparatus for implementing a linear transform on input data values to produce linear transformed output data, the apparatus comprising: input means for inputting input data values one after another to each of a series of multiplication means; a series of multiplication means interconnected with the input means for multiplying a current input data value by a constant to produce a current multiplier output; an interconnection network interconnecting the series of multiplication means to predetermined ones of a series of signed accumulator means; a series of signed accumulator means each interconnected to the interconnection network, each of the signed accumulator means producing an intermediate accumulator output by accumulating a corresponding one of the current multiplier outputs with a corresponding previous intermediate accumulator output, each of the signed accumulator means outputting the intermediate accumulator output as a corresponding linear transformed output data value.