Semantic Image Labeling with Histograms of Oriented Gradient and Gray Level Co-occurrence Matrix

In this paper, we propose a new approach for semantic image labeling by incorporating texture, gradient and color information. In our paper, the texture information is extracted by Gray Level Co-occurrence Matrix (GLCM). The gradient information is obtained by Histograms of Oriented Gradients (HOG). We apply the HOG, GLCM descriptors with color information simultaneously to enrich the image features of different information. To utilizing these features more effectively, we use the Approximate Nearest Neighbors (ANN) algorithm for clustering. After obtaining these information, the Joint Boost algorithm is applied to give an effective classifier by training many weak learner classifiers. At the end, a set of experiments with one descriptor or several descriptors combined are made to evaluating the performance of our method.