| Summary: | 碩士 === 逢甲大學 === 電機工程所 === 92 === The purpose of this paper is to construct a texture feature and image classificatiom query system, using the wavelet transform which decomposes image frequency to find some texture features in the JPEG 2000 of compression domain. Make use of the human special thinking creativity, the flexible characteristic recognition capability, and the fast cognition capability of the complicated graphics to make up for the inflexibility of computer algorithm, present vast and multiple data by graphics, and implicit information by visual pattern.
Make use of the mankind's gifted fast vision characteristic recognition capability to extract the useful information quickly. In light of this, the paper present a texture classification query system which uses the frequency division of wavelet transform to solve the problem of texture extraction, combine the genetic algorithms and classify the ethnicity with the relative topic on the image data in order to categorize the same image of veins in the same cluster and let users inspect the alike image data in each cluster more quickly.
The three essential parts in the paper are briefly described in the following:
(1) Texture extraction: First, make use the idea of the wavelet transform, and make use that three texture features the edge in the Hi- band are stronger, and election the sub- bands of the most texture features in the image. The experiment research that pays attention to manage the physics indicates that the mankind are most in common use to describe the most use elegant language of texture is directionality, coarseness and contrast, and know with the directionality feeling most severely, the next in order is coarseness, contrast . Therefore, we join the condition of the human visualization be the design fitness function. Defined the threshold, with texture of directionality for superior consider first, a time of coarseness, an end of contrast. Construct a total table of the texture features.
(2) Image classification by Genetics Algorithms (GA) : We develop a Genetic algorithm to classify the image database. The advantages of Genetic algorithm are searching in parallel and multi-points, therefore it always lead to a global optimal solution. Furthermore, we design a similarity evolution model in order to achieve a better performance of the retrieval system.
(3) Query by example: A query image is used as an sample for image retrieval. Since direct computation of distance is expensive, we develop a triangular inequality algorithm to eliminating most of the dissimilar images in database. The remain candidates are compared with the query image by the direct computation of distance. In this way, one can reduce a large amount of the distance computation during the image retrieval process which is urgently needed in large image database.
|
|---|
