ECG Signal Detection via Bidirectional Projection Clustering-enhanced Hashing

Abstract

Hashing has been extremely regarded in the field of machine vision due to its fast query ability and lightweight storage. Whereas, the binary discrete constraint invariably disturbs the optimization of various bit allocation algorithms. This article engineers a Bidirectional Projection Clustering-enhanced Hashing (BPCH) framework which dexterously avoids the binary optimization and can generate binary codes without any iteration. Concretely, BPCH leverages clustering to generate pseudo tags and class centers and takes their binary dissimilarity as the calibration, and then conducts Kronecker product with the real semantic affinity to generate a binary matrix, which is then rearranged according to the semantic tags to generate the final hash encodes. The whole process not only avoids the problem of binary optimization, but also does not involve any iteration, thus improving the robustness of hash book and reducing the impact of semantic information noise. Furthermore, a bidirectional projection hashing is constructed to link the raw data space and the latent Hamming space, thus providing a directly practicable hash function for out-of-sample data. Experimental results on two ECG data sets show its superiority over the current ECG signal retrieval algorithm.