Multiple Object Classification Using Hybrid Saliency Based Descriptors

We propose an Automatic approach for multi-object classification, which employs support vector machine (SVM) to create a discriminative object classification technique using view and illumination independent feature descriptors. Support vector machines are suffer from a lack of robustness with respect to noise and require fully labeled training data. So we propose a technique that can cope with these problems and decrease the influence of viewpoint changing or illumination changing of a scene (noise in data) named the saliency-based approach. We will combine the saliency-based descriptors and construct a Feature vector with low noise. The Proposed Automatic method is evaluated on the PASCAL VOC 2007 dataset.