Enhanced Surface Classification from Tactile Data by Image Fusion

Anwesha Khasnobish, Monalisa Pal, Amit Konar, Dewaki N. Tibarewala, and Atulya K. Nagar

Keywords

Tactile images, textures, surface classification, support vector machine

Abstract

This work deals with recognizing surface by processing local information from their corresponding tactile images and fusing them to obtain the global pattern of surface irregularities. Tactile images are acquired while exploring surfaces with four kinds of texture patterns. Texture information is obtained from each of the images by edge detection of the region where higher amount of pressure is felt. These edge detected images are fused to obtain the pattern of surface irregularities. The fused images are classified using hierarchical multi-class Support Vector Machine which yields an accuracy of 83.334% in 0.083 seconds. It is observed that the classification accuracy is enhanced by image fusion than that obtained by concatenating features of each component image which formed the fused images in the former case. When noise is gradually added to the features, the classifier shows an accuracy of 75% even when SNR is 8dBW, indicating the robustness of the classifier. Also, the performance of the algorithm is tested by adding white Gaussian noise to the raw images. Finally, McNemar Test validates the results. Thus, the algorithm can be integrated into a tactile-sensing system in real-time scenario for identifying surfaces based on texture.

Important Links:



Go Back