Concept Detection in Images Using SVD Features and MultiGranularity Partitioning and Classification

New visual and static features, namely, right singular feature vector, left singular feature vector and singular value feature vector are proposed for the semantic concept detection in images. These features are derived by applying singular value decomposition (SVD) directly to the raw images. In SVD features edge, color and texture information is integrated simultaneously and is sorted based on their importance for the concept detection. Feature extraction is performed in a multigranularity partitioning manner. In contrast to the existing systems, classification is carried out for each grid partition of each granularity separately. This separates the effect of classifications on partitions with and without the target concept on each other. Since SVD features have high dimensionality, classification is carried out with Knearest neighbor (KNN) algorithm that utilizes a new and stable distance function, namely, multiplicative distance. Experimental results on PASCAL VOC and TRECVID datasets show the effectiveness of the proposed SVD features and multigranularity partitioning and classification method