Interval Valued Feature Selection for Classification of Logo Images

Abstract

A model for classification of logo images through a symbolic feature selection is proposed in this paper. The proposed model extracts three global features viz., color, texture, and shape from logo images. These features are then fused to emphasize the superiority of feature level fusion strategy. Due to the existence of large variations across the samples in each class, the samples are clustered and represented in the form of symbolic interval valued data during training. The symbolic feature selection is then adopted to show the efficacy of feature sub-setting in classifying the logo images. During testing, a query logo image is classified as one of the members of three classes with only few discriminable set of features using a suitable symbolic classifier. For experimentation purpose, a huge corpus of 5044 color logo images has been used. The proposed model is validated using suitable validity measures viz., f-measure, precision, recall, accuracy, and time. The results with the comparative analysis show the superiority of the symbolic feature selection method with that of without feature selection in terms of time and average f-measure.