PROJECT TITLE :

Context-Aware Discovery of Visual Co-Occurrence Patterns - 2014

ABSTRACT:

Once a picture is decomposed into a variety of visual primitives, e.g., native interest points or regions, it is of great interests to get meaningful visual patterns from them. Conventional clustering of visual primitives, but, sometimes ignores the spatial and have structure among them, thus cannot discover high-level visual patterns of advanced structure. To overcome this downside, we propose to contemplate spatial and feature contexts among visual primitives for pattern discovery. By discovering spatial co-occurrence patterns among visual primitives and have co-incidence patterns among different varieties of features, our technique can better address the ambiguities of clustering visual primitives. We have a tendency to formulate the pattern discovery downside as a regularized k-means that clustering where spatial and feature contexts are served as constraints to enhance the pattern discovery results. A unique self-learning procedure is proposed to utilize the discovered spatial or feature patterns to gradually refine the clustering result. Our self-learning procedure is absolute to converge and experiments on real images validate the effectiveness of our method.