Feature Extraction Based On Statistical Thermodynamics Descriptors And Application In CBMIR

In the field of information retrieval, how to effectively understand and retrievelarge-scale databases and obtain the required information has become a valuable researchfrom the mass image.Therefore, it has been an important practical significance, exploring apractical content-based image retrieval system and finding efficient feature descriptionmethods.So far, none of the feature extraction method is the best. CBMIR usually require a highretrieval performance and more appropriate image features. It also has a more accurate imageunderstanding and professional medical knowledge. This article solves these problemsdrawing on the ideas of statistical thermodynamics. In this paper, it has been presented a classof image features based on statistical thermodynamics and the corresponding description ofthe way. The main contents of this article are as follows:(1) There is a certain similarity on research methods between digital image and material.They all need to solve the problem between the microscopic and macroscopic properties offeatures. This article explores the principles and methods of statistical thermodynamics tointroduce to image processing and related fields.(2) Proposed a new class of image features. They are the "image energy","imagequality","image temperature" and "image pressure", and so on. It has been demonstrated thatthey can be expressed medical image content information reasonable.(3) By doing experiments comparing this method with the traditional methods.Traditional methods include (GLCM, regional texture features, etc.) may find that they haveinformation overlap.This conclusion indicating that they extract similar information.(4) This article uses the PCA method allows multiple features together. Fusion featuresinclude statistical thermodynamic characteristics, GLCM, regional texture characteristics.Then apply them to the medical image retrieval system, we get a higher precision and recall.Using this method to the brain image retrieval, experimental results show that thismethod is effective.