IDENTIFYING COMPLEMENTARY CORNER DETECTORS FOR CORRECT IMAGE PIXELS CLASSIFICATION

Abstract

Classification of digital image content is mainly done by identifying low level image features such as corners and edges. The literature shows variety of algorithms for the identification of corner and non-corner pixels, important for objects’ identification and image segmentation. However, all of these algorithms produce different results for same data and therefore, suitable for limited applications. This paper proposes a hybrid solution of combining complementary corner detection algorithms to improve image pixels’ classification. This has been done by identifying the best detection algorithm for corner points with small and large angles and producing a hybrid algorithm by combining the latter two. Results have shown that Harris detector combined with Global and Local Curvature Points (GLC) improved the detection rate by 28% in synthetic images, but 50% in real images whereas, the combination of Shi’s detection algorithm with GLC enhanced the detection rate by 25.9% in synthetic images and 123% in real images, showing a significant improvement.