Improved Material Recognition Using Neural Networks

Material recognition is a significant subtask in computer vision that has many different applications like autonomous cars in transport engineering, auto-lifting in robotics and even for some sort of autonomous-surgeries in medical field. In this paper, we suggest an improved method to recognize the materials of different categories from one photo image that is captured in an unidentified illumination and view-angle. For that purpose, several different features that cover dynamic material appearance in detail and all gathered information will be used by neural network for recognition. Our proposed method significantly improved in contrast to several leading alternatives including SIFT, Color SIFT, LBP, k-mean, and others. Additionally, we use a material database of Flickr Material Database ( FMD) to analyze our method for identifying materials of different categories including foliage, plastic, wood, glass, leather, metal, paper, fabric, water and stone.