Abstract:

This study provides a method of improving characteristics by selecting the character data obtained by the multi-resolution analysis method for the classification of mammography images. The process of selection of characteristics is based on the determination of the characteristics that most reveal the difference between classes. First of all, the images are divided using wave atom conversion and a character vector is created from the wave atomic ratioes obtained. The matrix lines correspond to the images, the columns to the character data. The method is to investigate columns (optimum properties) that represent class differences at the highest level with the help of marginal values by dealing with each column separately. In order to generalize the obtained optimal characteristics, the classification process is re-executed with 5 layer cross-verification. The results indicate that the assembly of characteristics obtained by the recommended method has sufficient ability to classify mammographic images.

Keywords: