Medical Image Retrieval Based On Nonsubsampled Contourlet Transform and Fractal Dimension

A novel medical image retrieval algorithm based on texture information is proposed. The texture image retrieval based on fractal geometry is a commonly used method. However, it is inadequate only using fractal dimension to describe the texture. The nonsubsampled contourlet transform has the properties of multi-scale and multi-direction. Firstly, the nonsubsampled contourlet transform were done on original texture image, and then the fractal dimension of the transformed image was computed. The algorithm extracts fractal features with scale and orientation characteristics. To decrease the gap between high level concepts in human minds and low level features computed by computers, an improved SVM relevance feedback is introduced according to users´ intention. A database of CT images was retrieved by this algorithm. The result shows it can achieve a high precision of retrieval.