| Summary: | 碩士 === 中原大學 === 資訊工程學系 === 85 ===
Codework search is an important problem arising in vector quantization. Many fast algorithms, which ensure to generate the same encoding quality as the full search algorithm, have been proposed for the problem. A common key idea behind these algorithms is to use some well-disigned test conditions to determine whether the distortion between an input vector and a codeword needs to be actually computed. Since the time spent on deciding the truth of the test condition is much less than that spent on computing the distortion, the overall search time in general can be reduced significantly.
In this thesis, we first review three fast codeword search algorithms, namly the DT (double test), IP (integral projections ) [4], and ENNS (equal-average nearest neighbor search ) [3] algorithms, as well as their test conditions. Then, we build up the relationships betweem these test conditions by proving three important theorems. The three throrem simply that developing a class of hybrid codeword search algorithms by combining the test conditions is feasible. Therefore, seven new hybrid algorithms, namely the DTIP, IPDT, ENNSP, DTENNS, ENNSIPDT, ENNSDTIP and DTENNSIP algorithms, are developed and presented in this thesis.
The seven new codeword search algorithms as well as the original threes all have been implemented in C language. Several real images are used as test data to compare the efficiency among the algorithms. The simulation results indicate that the class of algorithms which consist of the test conditions used by the IP algorithm beat all the other algorithms. Overall, the IP and DTIP algorithms could run faster than the others.
|
|---|
