Masaya Fujisawa

A class of quasi-cyclic regular LDPC codes from cyclic difference families with girth 8

In this paper, we propose a class of regular LDPC codes from a cyclic difference family, which is a kind of combinatorial design. These LDPC codes have no 4-cycles, i.e., cycles of length 4. We clarify the conditions on which these codes with column weight 3 have no 6-cycles and discuss their minimum distance. Finally, we show the performance of the proposed codes with high rates and moderate lengths

A fast interpolation method for list decoding of RS and algebraic-geometric codes

In this paper, we present an efficient algorithm to find a Grobner basis of the ideal set of polynomials I(V) based on the Berlekamp-Massey-Sakata (BMS) algorithm, which gives another efficient method of giving the solution at the first stage of the Sudan algorithm (1997). We also show that the interpolation problem can be generalized to find a Grobner basis of the ideal I(V;M) which consists of polynomials having zeros (/spl alpha//sub i/,/spl beta//sub i/)/spl isin/V with some multiplicity condition specified by a set M (/spl isin/Z/sub 0//sup 2/) of integer vectors. This Hermitian type of interpolation problem takes a role in the improved version of the Sudan algorithm. A modification of BMS the algorithm can be applied to solve this problem. On the other hand, for list decoding of one-point AG codes our method can be adapted to find a Grobner basis of a relevant ideal.

A Construction of High Rate Quasi-Cyclic Regular LDPC Codes from Cyclic Difference Families with Girth 8

In this paper we propose a method of constructing quasi-cyclic regular LDPC codes from a cyclic difference family, which is a kind of combinatorial design. The resulting codes have no 4-cycle, i.e. cycles of length four and are defined by a small set of generators of codes with high rate and large code length. In particular, for LDPC codes with column weight three, we clarify the conditions on which they have no 6-cycle and their minimum distances are improved. Finally, we show the performance of the proposed codes with high rates and moderate lengths.

Generating Summary Videos Based on Visual and Sound Information from Movies

Vast quantities of video data are now widely available and easily accessible; because of the many videos that users encounter, video summary technology is needed in order to help users find videos that match their preferences. This study focuses on movies to propose a method for extracting important scenes based on visual and sound information, and verifies the degree of harmony of the extracted scenes. The video segments thus characterized can be used to generate summary videos.

Augmentation and decoding of compound-error-correcting codes

A compound error is any combination of burst errors with various burst lengths including random errors. The compound weight of any such error is defined as a kind of combinational metric which is a generalization of Gabidulins metric. First, we present a fast method for calculating the weight of any word based on a kind of trellis diagram. Extending the two-dimensional trellis diagram to a three-dimensional trellis, we propose a method of constructing codes having higher coding rate by augmenting any compound-error-correcting codes. Furthermore, we propose a decoding method for any such codes by modifying the three-dimensional trellis in augmentation.