Localizing artistic objects in frescos through contour based features

This paper presents a new contour based feature in the application of localizing artistic objects in frescos of Dunhuang Caves. The method treats contour curves of an image as a set of straight lines and constructs a descriptor for each line. In the construction of descriptors, in addition to the length and direction of the line itself, we consider the context information in the line space. We express the context of a line as the distance, length ratio and direction difference between the line and its K nearest neighbor lines. And similar to the method in [1], which constructs a score function for each interest point, we construct a score function for each line through an SVM classifier trained by our contour line features. In the localizing process, we calculate a score for each line detected according to the score function constructed and assign the score to each pixel on the line, and further get a score matrix with the same size as the original image. Then we apply ESS[1] algorithm to search the location of the artistic object according to the score matrix generated to complete the localization task. Experimental results show that our feature works fine in localizing artistic objects in frescos.