| Summary: | 博士 === 國立成功大學 === 電機工程學系 === 88 === Image coding methods are useful in reducing the storage and transmission bandwidth requirements of images through efficient data representation. Block Truncation Coding uses a two-level moment preserving quantizer that adapts to local properties of the images. It has the features of low computation load and less memory request while its bit rate is only 2.0 bits per pixel. Genetic algorithms are search algorithms based on the mechanics of natural selection and natural genetics. Recently, GAs had been used in the fields of image compression. Some new skills of GAs are developed in this Thesis.
In this Thesis, a new scheme of designing two-level minimum mean square error quantizer for image coding is first development. Genetic algorithm is applied to achieve this goal. Comparisons of results with various methods have verified, the proposed method can reach nearly optimal quantization with only less iteration.
A new scheme of designing multilevel BTC coding is proposed. The optimal quantization can be obtained by selecting the quantization threshold with an exhaustive search. However, it requires an enormous amount of computation and is thus impractical while we take an exhaustive search for the multilevel BTC. The two-steps searching and the genetic algorithm methods are applied to reduce the computational complexity. Comparisons of the results of the proposed methods with the exhaustive search reveal that the proposed methods can almost achieve optimal quantization with much less computation that that required in the latter case.
A color image compression method based on genetic algorithm and absolute moment block truncation coding is proposed. Color images comprise three planes, including red, green, and blue. There is very high correlation between the image in these planes. This motivates the use of one common bitmap to represent all three of the color bitmap. In order to generate such a bitmap so that the average mean square error between original and reconstructed images is a minimum, the genetic algorithm is applied. Comparison of results with various methods have verified that the proposed method has higher performance than the other schemes for single bitmap BTC coding of color images.
To further reduce the bit rate of two output data of AMBTC, we propose postprocessing methods to achieve this goal. The block of 2 x 4 bit-map is packaged into a byte-oriented symbol. The entropy can be reduced from 0.965 bpp to 0.917 bpp in average for our test images. The two subimages of quantization data (a, b) are post-processed by the Peano Scan. This postprocess can further reduce differential entropy about 0.4 bit for a 4x4 block. By applying arithmetic coding, the total bit reduction is about 0.3~0.4 bpp. The bit rate can achieve 1.6~1.7 bpp with the same quality of traditional AMBTC.
|
|---|
