Abstract: The methods of the invention can be used with image data divided into domain blocks. A predetermined search pattern of range blocks centered on a domain block is defined for use in the methods. The first method includes a step of generating at least one error descriptor data based on domain block data and range block data. The error descriptor data can be derived by scaling the range blocks to the pixel size of a domain block, and subtracting the means of the range blocks and domain block from each pixel thereof. The mean-adjusted, scaled pixel intensity levels of the scaled range blocks are subtracted from mean-adjusted pixel intensity levels of the domain block to produce difference data. The absolute value of the difference data is taken and the positive difference data are summed to produce summed error data for each range block. The summed error data is used to derive at least one error descriptor data for the image.

Abstract: The image reconstruction method generates higher quality reconstructed images from compression of a bi-level image. The method includes thresholding to force pixels of the starting image to be closer to bi-level to generate a threshold image. The method transforms the threshold image to generate transform coefficients representing the decomposition of the threshold image. The method selectively clamps the transform coefficients into quantization bins defined by compression of the bi-level image. The selective clamping generates modified coefficients corresponding to the higher quality reconstructed image. The method also includes applying an inverse-transform on the modified coefficients to generate higher quality reconstructed image. The starting image and the reconstructed image can be compared to determine the degree of improvement obtained with the method. The method can be repeated iteratively to obtain an image that more closely represents the original image than does the original lower-quality image.

Abstract: A method and apparatus are described for providing a scaled expanded image array of an original image array representing a physical entity, such as an image or sequence of frames, by utilizing fractal transform methods. The method of the invention includes steps performed by a data processor. Fractal transform methods are used to select a best domain/range block correspondence in an original image array. The range block is larger than the domain block. A subset of the data values in the range block is selected. Finally, an area of the expanded image array is populated with the data values in the selected subset where the area corresponds to the domain block location in the original image array.

Abstract: Missing intensity values for color values for a color component array derived from a color filter array are predicted using the system and method of the present invention. The method uses predictor sets for each missing intensity value to select a best value for each missing intensity value. The predictor sets include a luminance predictor set, a first chrominance predictor set and a second chrominance predictor set. The member predictors of the luminance predictor set are selected in response to a comparison between a variance between neighboring luminance values and a predetermined threshold. The first chrominance set is used to predict chrominance values of a color type at locations where a chrominance value of another color type was originally stored. The second chrominance set is used to predict chrominance values of a color type at locations where a luminance value was originally stored.

Abstract: A system and method is disclosed for representing each frame in a sequence of video frames in terms of a correlation between successive video frames. The system and method of the present invention segment the current video frame data and a previous frame into blocks for which motion vectors and/or affine mapping coefficients may be determined. The motion vectors may then be used to represent the current frame data without requiring periodic transmission of a base, self-referential frames of video data. Preferably, the compressor which generates the codewords to represent the current frame data subdivides the blocks of the current frame and previous frame buffers to further evaluate motion factors which may be used to represent current frame data. This evaluation is performed in a recursive manner and preferably includes evaluation of a bit rate/distortion factor to optimally select codewords for representing the current frame data.

Abstract: In one method of the present invention a plurality of thresholds ranges is selected. A first coefficient of the tree structure is compared against the plurality of threshold ranges during a first single traversal of the tree structure and a second coefficient of the tree structure against the plurality of threshold ranges during the first single traversal of the tree structure. The method of the present invention also determines whether the first or second coefficient satisfies a selected condition with respect to either of the threshold ranges. The difference in addresses between the addresses of the first and second coefficients is determined if the first and second coefficient satisfy the selected condition. The method of the present invention then codes the location of the second coefficient based on the address of the first coefficient using the difference.

Abstract: A method for compressing related data sets of a sequence is disclosed. The method compares domain blocks from a current frame buffer to range blocks in a previous frame buffer or vector quantization ("VQ") tables to generate codes. The method includes determining an initial codeword that represents a domain block of a current frame buffer in reference to a selected range block in a previous frame buffer. The selected range block is identified by a best measurement determined from an initial type of search performed on a first group of range blocks in the previous frame buffer. A cost factor is generated for coding the initial codeword. A first estimated cost of a codeword representing the domain block in reference to a second range block is determined. The second range block is identified by an estimated best measurement determined for a second type of search performed on a second group of range blocks.

Abstract: A method for transmitting fractal transform parameters generated by a variety of compressors using different fractal transform parameters is disclosed. The method includes generating an universal header having image reference data, color space data, fractal transform region reference data, and file decoding data and generating a fractal transform data segment having fractal transform parameters for domains and ranges used by a compressor to generate fractal transform parameters. The image reference data includes image dimensions expressed in physical size units so that the regeneration of the image at a decompressor is resolution independent. The color space parameters include identification of the original image color space and may include identification of a color space into which the original image is converted.

Abstract: A system and method for compressing and decompressing datastreams with fractal coding is disclosed. The method includes the generation of an optimal data set transformation function corresponding to a representative datastream captured from a data source. The data set transformation function is determined by ordering the digits of the representative datastream to order the intervals of an attractor. The ordered intervals are then used to determine the mapping functions which define an optimal data set transformation function. The nodes or graphical representation of the data set transformation function are incorporated in a compressor/decompressor combination. The compressor receives data symbols from a data source and determines the next interval of an attractor using a composition operation. The address or a portion of the address is shifted out as a number representative of the data symbols. Preferably, the interval is renormalized so the interval may be represented in sixteen bits.

Abstract: A system and method are disclosed for representing a data set by selecting a data transformation function and a data masking function. Preferably, the data set transformation and data masking functions are used to generate an attractor that more accurately represents the data set than an attractor generated by the data transformation function alone. The data masking function is a set of exclusionary data elements which are used to terminate data transformations for data elements which generate the attractor. Preferably, the data masking function is defined as a polygon for a two dimensional space. When the data masking function is incorporated with the data transformation function, the attractor normally produced by the data set transformation function is constrained to more accurately represent the data set.

Abstract: A method for encoding and decoding digital signals, particularly audio signals, by using multiple reference vectors from a codebook and adaptively updating the codebook. The reference vectors from the codebook are combined, using a mathematical combination function, to achieve a synthesized vector that most closely matches the digital signal. The encoded signal consists of the mathematical combination function coefficients and the memory addresses of the reference vectors from the codebook used to achieve the synthesized vector. The encoded signal is decoded by separating the coefficients and the addresses and using a reverse combination procedure to reproduce the synthesized vector. Finally, the synthesized vector is stored in the codebook, by overwriting the oldest data, to adaptively update the codebook.

Abstract: A method for encoding and decoding digital signals, particularly speech signals, by separating the original signal into a linear and nonlinear portions, the nonlinear portion may have linear aspects. The linear portion is encoded by LPC techniques and the nonlinear portion is encoded by use of a fractal transform. The encoded signal consists of LPC filter coefficients and fractal transform coefficients. The encoded signal is decoded by separating the LPC coefficients and fractal coefficients, using the LPC coefficients to generate an LPC filter, decoding the fractal coefficients using a fractal transform decoding method to obtain an error signal, and exciting the LPC filter with the decoded error signal to obtain decoded digital signals.

Abstract: A method of fractally interpolating intensity values for a color component array derived from a single chip charge-coupled-device (CCD) is disclosed. The method includes selection of a domain area about an array element not having a value for the color component array. A plurality of range areas are selected which are larger than the domain area. Additional range areas may be generated by range area transformations and intensity value scaling and shifting. A difference measurement between each domain area and range area is determined and the range area corresponding to the smallest distance measurement is selected. The intensity values of the selected range area are then used to generate an interpolated intensity value for the domain area.

Abstract: A system and method for interpolating intensity values for a color component array are disclosed. Intensity values stored in a color component array are derived from intensity data sensed by a charge-coupled-device in a camera. First and second gradients are determined for the array elements which have no color intensity value corresponding to the color component stored in a first color component array. The gradients are used to select intensity values from the first color component array and a second color component array. The second color component array includes the intensity values for the color component which was sensed by the array element having no intensity value in the first color component array. Incorporation of intensity values of both the first and second color components reduces the likelihood of color artifacts.

Abstract: A system and method compresses datasets so compressed representations corresponding to a predetermined target file size are generated. The system includes a lossy compressor which selects a compressed representation for components of an original data set. The size of the compressed representations generated by the compressor are measured and used with the cost parameters for a parameterized non-linear relationship between cost parameter and compressed representation sizes. The parameters may be used to adjust the cost parameter so the compressor generates compressed representations that correspond to a target size. The adjustment of the cost parameter causes the compressor to select different compressed representations for the components of the original dataset and correspondingly alters the size of the compressed representation.

Abstract: A system and method for contractively mapping decompressed images to facilitate resynchronization of data transmission is provided. This system includes a transmitter having a data compressor, a decompressor, and a contractive mapper. The receiver for the data transmission includes a decompressor and contractive mapper. The decompressor and contractive mapper of the transmitter provide the compressor with the same information which the decompressor at the receiver is using to decompress data. In this manner, the compressor may accurately represent the incoming frame data in terms of the contractively mapped decompressed data. The contractive mapper attenuates information in the decompressed data frame to prevent the persistence of erroneous data at the receiver should data become corrupted during transmission or the transmission be interrupted. The contractive mapping is preferably achieved with an affine map contracting scheme which is preferably implemented by a scalar multiplication.

Abstract: An encoder and decoder system for compressing discrete data sequences is disclosed. The encoder includes a self-affine map generator and a backward adaptive map generator which model source vectors formed from a discrete data sequence. The self-affine map generator uses known modeling techniques to determine map parameters for mapping functions. The mapping functions are evaluated by computing a collage error and the mapping function corresponding to the smallest collage error is selected. The backward adaptive map generator determines map parameters for mapping functions which map vectors previously generated to represent the discrete data sequence to the source vector. A distance measurement is computed for each target vector/mapping function combination and the mapping function corresponding to the smallest distance measurement is selected.

Abstract: A system and method for analyzing a set of data elements is disclosed. The method includes the steps of transforming data elements of a transform data space using the states of data elements comprising a data set and determining whether the transformed data elements have structure in the transform data space. The transformation of the data elements is performed by iteratively selecting mapping transformations corresponding to the states of successive data elements in the data set and applying the mapping transformations to successive transformations in the transform data space. By viewing the transformations in the transform data space, structure may be ascertained and evaluated to obtain information about the data set. The inventive system includes a transformer that changes state in response to the state of the next data element received from the data set and transforms the last transformation stored in the transform data space.

Abstract: Digital image data compression apparatus includes a controller circuit for receiving digital image data and for processing the image data into blocks. The controller circuit supplies processed image data to a plurality of transform circuits and to a feeder circuit. The transform circuits receive data from the controller circuit and the feeder circuit, and provide parallel processing to compare blocks of image data and generate fractal transform values representing the image data in a compressed form.

Abstract: Digital image data compression apparatus includes a controller circuit for receiving digital image data and for processing the image data into blocks. The controller circuit supplies processed image data to a plurality of transform circuits and to a feeder circuit. The transform circuits receive data from the controller circuit and the feeder circuit, and provide parallel processing to compare blocks of image data and generate fractal transform values representing the image data in a compressed form.