Hybrid video compression method

Method implemented on a computer having a processor and a memory coupled to said processor for compressing a digitally coded video frame sequence, comprising the steps of

• a, dividing a given frame into two-dimensional blocks,
• b, establishing a block partitioning of the frame, in specific cases by dividing individual blocks into further sub-blocks,
• c, carrying out on the information content of blocks a transformation (DCT) that converts spatial representation into frequency representation, producing thereby transformed multiple-element two-dimensional blocks and
• d, modifying the elements of the transformed blocks according to external boundary conditions, and finally
• e, encoding the information contained in transformed blocks by entropy coding,
• characterised by that
• in step d, the modification of the data in transformed multiple-element-two-dimensional blocks is modified depending on the size of the blocks and on the bandwidth available for transmitting coded data, and
• wherein at least some of steps a through e are performed using said processor.

Method implemented on a computer having a processor and a memory coupled to said processor for compressing a digitally coded video frame sequence, where the information content of certain frames is encoded from the contents of the preceding or subsequent frames (reference frames), the method further comprising the steps of

• a, dividing the frame to be encoded into blocks,
• b, searching a matching reference block for the given block to be encoded in the reference image preceding or following the frame containing said block to be encoded,
• c, carrying out a compressibility analysis by comparing matched reference blocks and the block to be encoded,
• d, selecting the best reference block relying on the result of the compressibility analysis, and
• e, encoding said block using the best reference block just selected,
• characterised by that in step b, during the search for reference blocks:
• i) the block to be encoded is divided into sub-blocks,
• ii) the contents of the sub-blocks are analysed,
• iii) according to pre-defined criteria, a predetermined number of sub-blocks, preferably at least two, are selected,
• iv) a reference block search is performed using the selected sub-blocks, said search being performed in a specific search range in the selected reference frame for the reference block containing sub-blocks that differ the least from the selected sub-blocks, with the relative position of the selected blocks kept constant during said search, and
• v) the best reference block is chosen as a result of a search using the selected sub-blocks,
• wherein at least some of steps a through e are performed using said processor.

Method implemented on a computer having a processor and a memory coupled to said processor for compressing a digitally coded video frame sequence, comprising the steps of

• a, dividing each frame into blocks that are to be separately coded,
• b, carrying out on the information content of the blocks a transformation (DCT) that converts spatial representation into frequency representation, producing thereby transformed blocks, and finally
• c, encoding the information contained in transformed blocks by entropy coding, and applying arithmetic coding as entropy coding, during which
• a bit sequence is encoded by modifying the lower and upper limit of an interval as a function of values of consecutive bits of the bit sequence, and
• the distribution of the already arrived bits of the sequence is taken into account in the function that modifies the limits of said interval,
• characterised by that
• addresses are generated from already arrived bit values of the bit sequence,
• said addresses are applied for addressing individual processing elements of a neural network comprising multiple processing elements, and
• parameters of the processing element are modified such that the frequency of individual addressing operations and the value of the currently arriving bit of the bit sequence are used as input data, and the output of the neural network is applied for determining a parameter that modifies the lower or upper limit the interval,
• after an initial learning phase involving the processing of multiple bits, the upper or lower limits of the interval being determined during the encoding of incoming bits as a function of the output of the neural network,
• wherein at least some of steps a through c are performed using said processor.

Method implemented on a computer having a processor and a memory coupled to said processor for compressing a digitally coded video frame sequence, comprising the steps of

• a, dividing a given frame into two-dimensional blocks,
• b, carrying out on the information content of blocks a transformation (DCT) that converts spatial representation into frequency representation, producing thereby transformed multiple-element two-dimensional blocks and
• c, modifying the elements of the transformed blocks according to external boundary conditions, and finally
• d, encoding the information contained in transformed blocks by entropy coding,
• characterised by that
• modification of the data of the transformed multiple-element two-dimensional blocks is carried out in step c, as a function of the output of a neural network,
• wherein at least some of steps a through d are performed using said processor.