Image quality measures for hierarchical decomposition of a shoeprint image

In an automated shoeprint classification and retrieval system, several practical difficulties exist hindering the effectiveness of shoeprint classification, such as device-dependent noise, distortions, and incompleteness. This makes it desirable to estimate the quality of a shoeprint image before it is fed into the process of feature extraction. It helps the system decide the types of image denoising, enhancement, and restoration required. Also, a hierarchical decomposition based on image quality measure can provide the position information for feature descriptors and the weights for different sections in feature matching. In this paper, we consider some first- and second-order statistics on shoeprint image quality estimation, and propose some gradient-based and ridgelet-based quality measures for the same purpose. According to their performance on the differentiation of ‘good’ and ‘poor’ shoeprint images alone, only eight of them are applied for shoeprint image quality estimation. Experiments on a database of ‘good’ and ‘poor’ shoeprints suggest that our approaches can provide a reasonable estimation of shoeprint image quality.