公开(公告)号：SG155797A1

公开(公告)日：2009-10-29

申请号：SG2008022717

申请日：2008-03-20

Applicant: ST MICROELECTRONICS ASIA ,  ST MICROELECTRONICS SA

Inventor： HUANG YONG ,  LEBOWSKY FRITZ ,  WANG HAIYUN ,  HUI LUCAS

Abstract: A video image processing system is described that generates the interpolated video images with sharp and jaggedness-free edges. A method of video image processing is also described that interpolates video images to generate the video images with sharp and jaggedness-free edges.

公开(公告)号：FR2830143A1

公开(公告)日：2003-03-28

申请号：FR0112209

申请日：2001-09-21

Applicant: ST MICROELECTRONICS SA

Inventor： BARD JEAN MICHEL ,  DANGER JEAN LUC ,  HUI LUCAS ,  CUNAT CHRISTOPHE

IPC: H04N7/26 ,  H04N7/30 ,  H03M7/30

Abstract: The method comprises the formation of spatial blocks of coefficients assigned at pixels, the transformation of each block into frequency domain, the association of a set of frequency blocks into macro-blocks and a set of macro-blocks into segments, and performing two steps of quantification for each segment. The first quantification of coefficients of each frequency block is with identical index for all frequency blocks of a macro-block. The second quantification of coefficients of at least a part of the frequency blocks is with a second set of quantification step substituting the first quantification step. The two sets of quantification step are chosen as a function, on one hand, of a target value (VC) of the quantity of data necessary for representing the quantified coefficients of the segment, and on the other hand, of gain factors computed for the frequency blocks of the segment. The gain factor relative to a block represents a reduction of the necessary quantity of data at the time of passage from one set of quantification step to another for the block. The index identical for all frequency blocks of a macro-block is also identical for all macro-blocks of a segment. The first set of quantification step is chosen so that the quantity of data representing the quantified coefficients of each data segment is higher than the target value. The second set of quantification step is coarser than the first; the quantity of data is less than or equal to the target value, and the quantification is effected in priority for the frequency block of the segment for which the gain factor is higher and repeated for at least another frequency block of the segment in a decreasing gain order as long as the quantity of data is higher than the target value. The gain factor computed for a frequency block is the difference between the quantities of data necessary for representing the quantified coefficients of the frequency block with the first set and with the second set, and divided by the quantity of data necessary for representing the quantified ceofficients with the first set. The first quantification comprises three processings in parallel, each comprising the operations of quantification, coding and computing the quantity of data necessary for representing the quantified data of the segment. The quantification is performed by use of an algorithm wherein a parameter (QID) is initialized to zero (120), the length (LS) of compressed data is computed (130), a test is performed if the length (LS) is less than or equal to the target value (VC) (140); in the case the test is negative, the parameter (QID) is incremented by unity (150) and the computing is repeated; the operations (130,140,150) correspond to the first quantification; if the test is positive, a test (160) verifies if the parameter (QID) is equal to zero; if it is indeed the case, the data cannot be more compressed and no further processing is required; if the test is negative, a value of the parameter (QID) decreased by unity is associated with all blocks of the segment (170), a permutation (sigma) of blocks is performed (180) in the order of decreasing quality factor, the index (j) is initialized to zero, the parameter is incremented by unit (200), the length (LS) is computed (210) and a test (220) is effected if the length (LS) is less than or equal to the target value (VC); if the test negative, the index (j) is incremented by unit (230); the operations (200,210,220,230,240) correspond to the second quantification. A device is claimed that implements the method.

公开(公告)号：FR2830143B1

公开(公告)日：2004-02-27

申请号：FR0112209

申请日：2001-09-21

Applicant: ST MICROELECTRONICS SA

Inventor： BARD JEAN MICHEL ,  DANGER JEAN LUC ,  HUI LUCAS ,  CUNAT CHRISTOPHE

IPC: H04N7/26 ,  H04N7/30 ,  H03M7/30

Abstract: The method comprises the formation of spatial blocks of coefficients assigned at pixels, the transformation of each block into frequency domain, the association of a set of frequency blocks into macro-blocks and a set of macro-blocks into segments, and performing two steps of quantification for each segment. The first quantification of coefficients of each frequency block is with identical index for all frequency blocks of a macro-block. The second quantification of coefficients of at least a part of the frequency blocks is with a second set of quantification step substituting the first quantification step. The two sets of quantification step are chosen as a function, on one hand, of a target value (VC) of the quantity of data necessary for representing the quantified coefficients of the segment, and on the other hand, of gain factors computed for the frequency blocks of the segment. The gain factor relative to a block represents a reduction of the necessary quantity of data at the time of passage from one set of quantification step to another for the block. The index identical for all frequency blocks of a macro-block is also identical for all macro-blocks of a segment. The first set of quantification step is chosen so that the quantity of data representing the quantified coefficients of each data segment is higher than the target value. The second set of quantification step is coarser than the first; the quantity of data is less than or equal to the target value, and the quantification is effected in priority for the frequency block of the segment for which the gain factor is higher and repeated for at least another frequency block of the segment in a decreasing gain order as long as the quantity of data is higher than the target value. The gain factor computed for a frequency block is the difference between the quantities of data necessary for representing the quantified coefficients of the frequency block with the first set and with the second set, and divided by the quantity of data necessary for representing the quantified ceofficients with the first set. The first quantification comprises three processings in parallel, each comprising the operations of quantification, coding and computing the quantity of data necessary for representing the quantified data of the segment. The quantification is performed by use of an algorithm wherein a parameter (QID) is initialized to zero (120), the length (LS) of compressed data is computed (130), a test is performed if the length (LS) is less than or equal to the target value (VC) (140); in the case the test is negative, the parameter (QID) is incremented by unity (150) and the computing is repeated; the operations (130,140,150) correspond to the first quantification; if the test is positive, a test (160) verifies if the parameter (QID) is equal to zero; if it is indeed the case, the data cannot be more compressed and no further processing is required; if the test is negative, a value of the parameter (QID) decreased by unity is associated with all blocks of the segment (170), a permutation (sigma) of blocks is performed (180) in the order of decreasing quality factor, the index (j) is initialized to zero, the parameter is incremented by unit (200), the length (LS) is computed (210) and a test (220) is effected if the length (LS) is less than or equal to the target value (VC); if the test negative, the index (j) is incremented by unit (230); the operations (200,210,220,230,240) correspond to the second quantification. A device is claimed that implements the method.