Abstract:

Irregular regions are difficult to match directly, we therefore propose a novel matching algorithm, called the distance transform irregular region descriptor (DTIRD) based on distance transform. Firstly, in order to avoid ellipse fitting errors and the matching error of an estimated reference orientation, we determine the support region of irregular regions utilizing the    texture information both near the region boundary and in the region. Secondly, we give a division method based on distance transform, which can reduce the matching error caused by image intensity changes. Finally, since regions have different shapes, sizes and different pixel points, we use the mean standard deviation (MS) algorithm to construct the descriptor. Experimental results shows that the proposed algorithm has better matching performance for images with rotation, illumination change, blur change and JPEG compression change.