Granular Computing And Its Application In Image Segmentation

Granular Computing studies problems which are imprecise, partial true, fuzzy and unnumbered, based on different levels of granules. It abstracts basic principle via people's experience of problem solving, and it explores answers in endurable extent to make problems manageable, robust, costless, and harmonious. It covers all theories, methodologies, techniques, and tools which make use of granules, which is superset of rough set, fuzzy set, computing with word, quotient space and interval computing, and is subset of granular math. It is complement and development of traditional math which requires precise answer. It is also one of the hotspot of artificial intelligence. The correlative theories of it are all aim at special problems or special environment. There are nearly no intersection among them. Granular Computing tries to conform these theories and erect further theory.Rough set theory is one subset of Granular Computing. It is a kind of mathematic tool in studying imprecise, uncertain problems and it mainly solves approximate problem of information granules. Information system's reduction is a NP-hard problem and is a bottle-neck problem rough set theory. This paper studies many kinds of reduction algorithm, and analogies them, improve them to reduce its complex degree and make it flexible. Then the paper mixes granular computing and rough set theory, and expands a whollynew kind of algorithm------ granular reduction algorithm. The algorithmreduces the information system by bit granular matrix which is a new conception and tool. Then get the best decision rule. The algorithm is fit to both consistent and inconsistent decision table, and also it should be in virtue of tools to carry out the operation, because it takes matrix as tool. So the speed is quicker, and the efficient is higher.Image segmentation abstracts object from the whole image. It is bridge from image processing to image analyzing. So it is quite critical. In one side, it is base of object expression, which affects the measure of character heavily. In another side, it translate originality image into more abstract and more compact form, which makes further image analyze and image comprehend possible.Image segmentation solves approximate problems of information granules, while traditional algorithm solves mathematic problems. So the effect is not good if use traditional algorithm in image. In recent years, rough set theory was used in image technology and got some success. But nearly there is no method which is fit to color image.Based on thought of Granular Computing -----jumping among differentlevels in problem solving, the paper erected a new frame of color image segmentation.. And it also makes use of the result above: granular reduction algorithm. Granular Computing studied problems in different granules from different granule levels to cut account down by any possible. According to the required precision, image should be divided into average pieces or different pieces. Make knowledge reduction to every piece of image by granular reduction algorithm. The reduction result can be decision rule. Then the segmentation should be carried out using the rule and space information. In image system, granule-reduction algorithm takes image as information system which pixel is condition attribute and color class isdecision attribute.The paper took matlab7.3 as tool to write granule-reduction algorithm and color image segmentation program. The emulation result of the typical image in image technology shows the segmentation impression is good.In a word, Granular Computing abandons traditional thinking mode which describes problem accurately, but studies problems of uncertain, fuzzy, unnumbered ones with acceptable approximate answer. It makes the information processing more scientific, economic. The paper mixed Granular Computing, rough set theory and image segmentation in a system and it showed comprehensive character and huge vitality of Granular Computing. But this is just primary study. How to perfect the theory is also in study. Aim to concrete problem of this paper, the perfection of the algorithm and program is something needs to solve.