Classification of deformable and non-deformable surfaces by tactile image analysis

Touch sensation provides various information of the object under exploration including its shape, size, deformability, softness, and texture. In addition to shape, surface deformability is a very important and unique characteristic of a surface according to which we can distinguish different surfaces during tactile exploration. In this study our objective is to recognize various surfaces depending upon surface softness from their tactile images merely by processing single tactile images of each surfaces rather than dynamic tactile exploration of the surface. Tactile images of six surfaces of different surface deformability are considered. Tactile images are preprocessed, followed by feature extraction and two-level classification using linear discriminant analysis (LDA) and k nearest neighbor (kNN). The overall mean classification accuracy with LDA in two levels is 94.41%, with a mean classification accuracy of 95.50% and 93.33% in level 1 and 2 respectively. In both cases LDA performed better with a run time of 0.07 sec and 0.08 sec in two levels respectively. The two-level classification technique performed classification in a short time with reduced the misclassification rate. Thus this process can be implemented for efficiently surface classification in real time applications.