Complex sport image classification using spatial color and posture context descriptors and neural classifiers

Searching a required image based on the content and visual meaning of the image is challenging. One of the complex groups of images is sport image since a sport image may contain various relevant information such as players´ postures, textures and color of players´ clothes, and complicated background ground. Achieving high recognition rate depends upond the features extracted from the image. In this paper, a new set of features is introduced. The features are 64-bin color histogram, single valued decomposition matrices for Cb and Cr color spaces, and 2-dimensional discrete cosine transformation. Regardless of any neural classifiers, our proposed features gives higher accurate results when compared with the other proposed features, i.e. edge histogram (EH) and region-based shape (RS).