A novel approach of Content Based Medical Images Indexing System Based on Spatial Distribution of Vector Descriptors

In this paper, we propose a Content Based Medical Images Indexing System Based on Spatial Distribution of Vector Descriptors. The query image can be a portion of images, or entire ones. Our system is based on retrieving and classifying images by extracting local image features using descriptors and statistical distance measurements to compute similarity. For pattern recognition, we applied the approach of Scale Invariant Feature Transform (SIFT) descriptor which extracts the local features vector for the key-point in order to obtain a cloud of points indicating the locations of the key-feature. Then, we used the Quick Hull algorithm to delimit this cloud of points into a vector forming a pattern. The medical image dataset we used is provided by EL KASSAB hospital, Tunis, Tunisia. Our system was successfully tested on brain tumor MRI, and we are in the process to extend it to bone images.