Combination of SVM and FERN for GPU-assisted texture recognition on mobile devices

Abstract

Feature point matching and texture recognition are two of the most important problems in the image processing. Recently, several new approaches to these problems using simple local features and semi-naive Bayesian classification scheme have been developed. In our paper, we show how to enhance these techniques further by combining them with Support Vector Machines using online learning techniques. The resulting algorithm is simple, robust and can be adapted to various tasks in image processing. Furthermore, we demonstrate the advantages of our method by using it to achieve real-time texture recognition on a mobile device by utilizing parallel processing capabilities afforded by the device GPU.