International Journal of Computer Science Issues

More than a traditional journal...

This paper presents a comparative study of dimension reduction methods combined with wavelet transform. This study is carried out for mammographic image classification. It is performed in three stages: extraction of features characterizing the tissue areas then a dimension reduction was achieved by four different methods of discrimination and finally the classification phase was carried. We have experimented for this purpose K-Nearest neighbours classifier. Results show the classification accuracy in some cases has reached 100%. We also found that generally the classification accuracy increases with the dimension but stabilizes after a certain value which is approximately d=60. We also present the results as a projection of onto a two-dimensional space. In some cases we observed a clear separation between normal images and abnormal ones.