Efficient contour match kernel

摘要: Abstract We propose a novel concept of asymmetric feature maps (AFM), which allows to evaluate multiple kernels between query and database entries without increasing the memory requirements. To demonstrate advantages AFM method, we derive an efficient contour match kernel – short vector image representation that, due maps, supports scale translation invariant sketch-based retrieval. Unlike most short-code based retrieval systems, proposed method provides localization in retrieved image. The efficiency search is boosted by approximating 2D via trigonometric polynomial scores 1D projections. projections are special case AFM. An order magnitude speed-up achieved compared traditional polynomials. results image-based average expansion approach and, any learning, significantly outperform state-of-the-art hand-crafted descriptors on standard benchmarks. Our competes well with recent CNN-based approaches that require large amounts labeled sketches, images sketch-image pairs.