Recognition of Object Categories in Realistic Scenes

Abstract

Classification of images maps the image content into a certain semantic term such as categories, domain, object. Image classification should be able automatically check the existence of certain object (e.g. car, animal, and scene) in the image content. This task is still challenging in computer vision since we have to deal with the realistic image. The objective of this works is to discover the image classification methods by mixturing the existing techniques with the aim of the best results in classification. In this work, we implemented sparse coding method with spatial pyramid matching to classify the images. Beside gray SIFT, four SIFT color descriptors were also used as a local descriptor. Linear Super Vector Machine (SVM) is conducted for training and testing the images. The result of this work has shown that color descriptors improve significantly the classification rate compared to gray SIFT.