Vector quantisation index compression based on a Coding Tree Assignment Scheme with Improved Search-Order Coding algorithms

Abstract

This study proposes a Coding Tree Assignment Scheme with Improved Search-Order Coding algorithms (CTAS-ISOC) to enhance the coding efficiency of the original SOC by exploiting the correlations of the neighbouring blocks using the left-pair and upper-pair patterns in the index domain. The essential techniques consist of three major elements: the Neighbouring Index Code Assignment (NICA), the Left-pair Search-Order Coding (LSOC) and the Upper-pair Search-Order Coding (USOC). The NICA approach assigns a short code to the current index by using the statistics on the indices of the neighbouring blocks. The LSOC (USOC) compares the current left (upper) index pair with previous index pairs in a predefined search path. Since the predefined search path is exploited with a correlation viewpoint, both LSOC and USOC achieve better compression than the original SOC. Experimental results demonstrate the effectiveness of the proposed scheme in comparison with some existing popular lossless index coding schemes

Index compression for vector quantisation using modified coding tree assignment scheme

“…Thus, the MCTAS scheme is presented to improve the coding efficiency of the image VQ scheme and reduce time complexity of CTAS‐ISOC [29]. In MCTAS, the coding scheme used in [29] is modified and the dynamic index table coding (DITC) technique is proposed to improve the time complexity of ISOC [29] by exploiting the correlations of the neighbouring blocks in the previously constructed index table that is dynamically classified and updated for each current index. The searching matched index in the previously qualified index table rather than in the original vector‐quantised index map results in low‐time complexity.…”

“…Since a small block size is usually used in VQ and similar patterns may appear in different shapes, the index pair also has a high probability to appear in the neighbouring regions. Thus, a left search pair (LSP) is defined as a row vector V sp = [ I L I C ] and the upper search pair is defined as a column vector [ I U I C ] T (where ‘ T ’ denotes the matrix transpose operation) are adopted by ISOC [29] for matching indices. The DITC scheme extends the idea of ISOC by matching indices pair by pair.…”

“…The searching matched index in the pre‐qualified index table rather than in the index map speeds up the coding efficiency. Since the search orders of SOC [20] and CTAS‐ISOC [29] are clockwise and outward from the search centre I C , the search regions for encodings of the previous I C , current I C and next I C by SOC and CTAS‐ISOC are highly overlapping. Thus, redundant computation is inevitable for finding matched indices.…”

“…The evaluation of computation complexity of MCTAS is similar to CTAS‐ISOC [29]. If T NICA , T L , T U and T ORI are used to denote the average computation time required to encode an index using NICA, DITC with the left index pair, DITC with the upper index pair and the original index rule, respectively, and the operations of T L are similar to T U (compare [29]), the total computation complexity of MCTAS is where N NICA , N L , N U and N ORI are the numbers of indices encoded by NICA, DITC with the left index pair, DITC with the upper index pair and the original index, respectively. Since the NICA coding scheme is exploited to check if the current index is the same as one of its neighbours and the original index rule encodes the current index with a flag code plus its original index directly, it is obvious that T NICA and T ORI are much less than T L .…”

“…In the paper, the modified coding tree assignment scheme (MCTAS) is presented to improve the coding efficiency of the image VQ scheme and reduce time complexity of the CTAS‐ISOC [29] technique by exploiting the correlations of the neighbouring blocks in the previously constructed index table. Experimental simulations show that the proposed scheme reaches better performance in bit rate than other popular existing lossless index compression methods, such as SOC [20], MSOC [21], LCIC [22], ICS [25], AICS [26], ISAIAL [27], modified ISAIAL [28] and CTAS‐ISOC [29].…”