Sketch-Based Image Retrieval Through Hypothesis-Driven Object Boundary Selection With HLR Descriptor

摘要: The appearance gap between sketches and photo- realistic images is a fundamental challenge in sketch-based image retrieval (SBIR) systems. existence of noisy edges on key factor the enlargement significantly degrades performance . To bridge gap, we propose framework consisting new line segment -based descriptor named histogram relationship (HLR) noise impact reduction algorithm known as object boundary selection HLR treats extracted series piece-wise segments captures them. Based HLR, aims to reduce by selecting shaping that best correspond boundaries. Multiple hypotheses are generated for descriptors hypothetical edge selection. formulated find combination maximize score; fast method also proposed. distraction false matches scoring process, two constraints spatial coherent aspects introduced We tested proposed public datasets dataset three million images, which recently collected SBIR evaluation purposes. compared with state-of-the-art (SHoG, GF-HOG). experimental results show our outperforms Combined algorithm, improves performance.