Property On The Guidance Contains A Feedback Mechanism, The Content-based Image Classification

With the overarching development of Internet retieval industry these years, limitation of attention on website retieval has ready been transended in terms of retrieval content. Retievals of such complicated imformation as music, image, vedio are constantly popping up, which leads to the extensive application of image retieval technique. At reasent, since the the text-based picture rerieval technique, which relies on keyword as a clue with less resource spending, is adopted in all the mainstream search engines, content-based retieval picture technique with more preferable search results is more adaptive to the demand of internet image retieval development and has pratical significence. This paper realizes the cost effective pseudosemantic content-based image retrieval system, which is called Internet-oriented Picture Retrieval Technique(IPRT) in this paper.Attribute theory proposed by Prof. Feng, is a method of artificial intelligence under the guideline of rule of quantitive and quantitive change in dialectic materialism. This theory aims at finding out the essence of object via its qualititive and quantitive attribute and put forward a series of methods of pattern recognition, machine learning, evaluation and decision-making ect. on the basis of qualititive mapping. This paper uses attribute theory as reference when building feature extraction algorithm of IPRT system and applies the "analysis approach of attribute coordinate" into the retrieval of user's psychological preference.The main aspects dealt with this paper are structured as follows:1. Proposing the concept of "pseudosemantic retrival". Designing and realizing the completed IPRT system, a kind of CBIR system focusing its retrieval on the integration of four features as colour, texture, shape and sketch with preferable accuracy, higher coverage and faster execution speed.2. Improving the image feature extraction algorithm. Adopting the HSV colour space, which is more fitted to ocular image, to express the process of colour, texture, shape and sketch extraction. Dividing the colour space into levels of 18×3×3, which not only contains the colour imformation but also has fewer dimensions than the traditional 256-level gray expression. The complexity of this feature extraction algorithm in controlled within O(M×N).3. In terms of feture extraction of colour, shape and sketch, proposing the "three-level sketching method" and " difference algorithm on simple block level", which have preferable running efficiency and robustness.4. Applying the "analysis approach of attribute coordinate" in "attribute theory" to record user's behavior in the process of retrieval and distill the user's psychological preference, which can provide more satisfactory retrival results for the users.5. Introducing the feedback machanism for psychological preference learning, which makes the retrieval results steadily approach user's expectation.