Nonlinear hash image retrieval method based on equivalent continuous change binary discrete optimization

Abstract

A nonlinear hash image retrieval method based on equivalent continuous change binary discrete optimization comprises the following steps of: 1, converting a binary discrete optimization problem obtained after quantification of a general hash model into an equivalent continuous optimization problem, and acquiring a corresponding optimal solution; 2, randomly selecting a part of images from a to-be-searched image set D to form a training image set T; 3, constructing a nonlinear hash function through using a three-layer full-connection network; 4, acquiring a nonlinear hash objective function byquantizing a similarity preserving loss function and adding discrete orthogonal constraints and bit balance constraints; 5, optimizing the objective function, optimizing network parameters by using astochastic gradient descent method, optimizing a binary code by using the method in the step 1, and giving convergence analysis; 6, training the hash function; and 7, calculating a hash code to carryout image retrieval. The nonlinear hash image retrieval method is small in accumulative error, the continuous optimization problem and the binary discrete optimization problem are equivalent, optimization convergence is achieved, and retrieval precision is high.

Description

technical field [0001] The invention relates to big data processing and analysis in the field of big data and image retrieval in the field of computer vision, and is especially used in discrete optimization methods in hash learning and image retrieval in hash learning. Background technique [0002] With the rapid development of information technology and the promotion of big data technology, the Internet, mobile phones, and logistics networks generate massive image data every day, and the subsequent storage and transmission requirements make traditional image retrieval technology unable to adapt to large-scale In image search, efficient retrieval methods brought about by fast and compact feature representations have been extensively studied. [0003] As a representative method of fast and compact feature representation, hash learning has been widely used in large-scale image retrieval due to its high efficiency and small storage space. Hash learning mainly uses machine lear...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the shortcomings of the existing hash learning image retrieval optimization method that the cumulative error is large, the continuous optimization problem and the binary discrete optimization problem are not equivalent, and the optimization does not converge, the present invention provides a continuous optimization method with a small cumulative error Nonlinear Hash Image Retrieval Based on Binary Discrete Optimization with Equivalent Continuous Variation Equivalence and Optimization Convergence of Problem and Binary Discrete Optimization Problem

Images