Integral Clustering and spectral Feature Modeling for Sketch-based Image Retrieval

Image retrieval using free hand drawn sketch image is addressed in the presented paper. As the volume of image information's are rapidly increasing, retrieval of information from the image data set is observed to be a tedious task. Majority of image retrieval are developed using an image querying system, however, unknown images which is given input as a free hand sketch is less addressed. The recent developments in the sketch-based image retrieval (SBIR) were developed using spectral image representation. However, the problem to the existing SBIR approach using random distribution of feature set builds a larger search overhead and the diverse distribution leads to misclassification. The objective of the proposed work is to develop a faster sketch-based image retrieval system with higher accuracy. This paper presents a new clustering-based approach for spectral feature distribution using cluster gain and information selection approach. The relative distribution and the information gains are considered in the development of the proposed approach for classification. The proposed approach observes a higher retrieval accuracy and classification performance in comparison to the existing spectral feature classification.