Medical Image Retrieval by Region Based Shape Feature For CT Images

Content Based Image Retrieval (CBIR) is the huge field for image retrieval from the wide databases. It is one of the popular techniques from computer vision domain. CBIR consist of feature extraction followed by similarity comparison. Shape feature extraction can be made on two descriptors one is region based and other is contour based. This paper implements shape feature extraction with region based descriptors. Zernike moments and Hu's seven moments have been used as a feature extraction techniques and Support Vector Machine (SVM) is used as a classifier. Different distance metrics are then used for similarity comparison with these feature extraction methods for efficient results. For performance evaluation distance metrics used are Euclidean, Chebyshev, Cityblock, Canberra, Standardized Euclidean (Seuclidean). Medical database with 6 classes consist of 100 images each namely head, hip, shoulder, pelvis, knee, ankle is used. After obtaining all the experimental results in terms of precision and recall, a comparative study is made for selected database.