Salient Map Creation and Database Retrieval Based on Salient Features

Features of an image can be extracted with the help of salient maps. Feature extraction can be either in spatial domain or in transform domain. In this work, a novel salient map creation model based on wavelet transform domain. At first, the wavelet transform is employed to create the multi-scale feature maps which can utilise low-level features. This saliency detection model is based on high-pass coefficients of the wavelet decomposition after eliminating some high-frequency components of the image. The idea is to create the feature maps by Inverse Wavelet Transform (IWT) on the multilevel decomposition. Then, a model for the saliency map from these features is proposed. Based on these saliency maps common saliency from multiple images is extracted and the matching images based on input images are retrieved. For this matching purpose weakly supervised learning technique will be used. Support Vector Machine (SVM) is actually designed for binary classification. In this paper, Multi-SVM is used for one against all