Optimizing Non-Differentiable Metrics for Hashing

Image hashing embeds the image to binary codes which can boost efficiency of approximately nearest neighbors search. F-measure is a widely-used metric for evaluating performance methods. However, it non-differentiable and hence has not been used as an object function hashing. Heuristic algorithms, e.g. evolutionary computation particle swarm optimization (PSO), are good at optimizing objectives, while they inefficient in very high-dimensional variables commonly models. To address this contradict, we propose scheme bridge methods objective using PSO. The generate real-valued images then parameters quantization procedure optimized by Our incorporate wide range methods, heuristic algorithms metrics. Experimental results demonstrate that our be further improve existing