A Classification Method of Diagnostic Image based on Tangent Distance.pdfVIP

A Classification Method of Diagnostic Image based on Tangent Distance Tan Rui, Xuan Guorong Compute Science Department , Tongji University, Shanghai, China, 200092 Abstract In this paper we present a classification approach based on tangent distance, which is specially designed for many applications of image recognition, and discuss the application in our classification of radiographs. In contrast to the traditional method to determine tangent vectors, we design a systemic approach using the single value decomposition to extract the main changed characters of a certain image set. With this new method we achieved considerable results in different image sets including diagnostic images and successful application systems have been developed in practice. Keywords: IRMA, tangent distance, single value decomposition 1 Introduction In many application systems of digital image classification and recognition, such as diagnostic image classification, license plate recognition, printed character classification, there are so many uncertain factors in the pre-processing (screening, scanning and so on) that the images have been distorted. These transformations such as rotation, scaling, shear are known as affined transformations. Though these influence can be lessened by per-procession, there exist all kinds of difficulties in the practical applications. The traditional classifier based on Euclidean Distance can not solve the problem of affine transformations. In contrast, the classifier based on Tangent Distance can avoid the influence of distortion effectively. The radiographs of the IRMA database mentioned in [3], which consists of 110 abdomen, 706 limbs,