Discriminative Directional Texture and Color for Robust Target Tracking

An efficient algorithm is proposed that enhances the traditional mean-shift color histogram based object tracking approaches. We propose prominent local directional pattern variance (LDPT) that can extract directional texture responses and with color we develop a color-texture histogram representation for the target. Experiments show superior performance that allows target to go under complex environment like illumination change, similar background color on which traditional methods fail.