Robust Local Descriptor for Color Object Recognition

Received: 26 August 2019 Accepted: 13 November 2019 Image category recognition is important to access visual information on the level of objects and scene types. In this paper, we propose a new approach for color object recognition using the powerful information provided by the color. This approach is based on the combination of Gray-Edge color constancy, hue components in HSV (hue, saturation, value) color space and cell and bin ideas used in the HOG (Histograms of Gradients) descriptors. The proposed oriented descriptor benefits of the invariance of hues against light intensity change, light intensity shift and light intensity change and shift, and solve its missing of invariance against light color change by using Gray-Edge color constancy. Moreover, the use of cells and bins in this proposed descriptor building boost its invariance the geometric and photo-metric transformation and increases the recognition rate. SVM classifiers (Support Vector Machine) which is a strong classification method known for its flexibility and its power of generalization are used for the training and recognition steps. The proposed method is evaluated on two publicly available datasets including Columbia Object Image Library and The Amsterdam Library of Object Images and obtained a recognition rate of 95.64% and 96.48% – clearly showing the exceptional performance compared to existing methods.