Data processing method for sketch retrieval

Abstract

The invention provides a data processing method for sketch retrieval. A knowledge preserving mechanism is introduced; on the basis of a domain loss function, a triple loss function and a semantic lossfunction, a knowledge preserving loss function is additionally constructed, and all the loss functions are used for training; in the knowledge migration process of the pre-training model, new knowledge of a target domain can be learned, knowledge learned from a source domain can be kept, and sketch retrieval precision is improved.

Description

technical field [0001] The invention relates to the field of sketch retrieval. In particular, it relates to a data processing method for sketch retrieval. Background technique [0002] At present, there are two main retrieval methods, one is text-based image retrieval, users search for images with the same or similar tags through keyword search; the other is content-based image retrieval, users search for the same or similar content by inputting images Additional pictures of . Among these retrieval methods, with the use of touch-screen devices becoming more and more common, hand-drawn sketches are convenient and fast, so the technology of using sketches to retrieve images shows great potential. Sketch retrieval aims to retrieve pictures similar to it using simple line outlines. Using deep learning technology to build a retrieval model requires a large number of data samples. However, the image information in the world is very rich, but the sketch materials are relatively ...

AI Technical Summary

Problems solved by technology

However, after knowledge transfer, although the classification ability of the obtained Siamese network on the target domain is continuously improved, the classification ability on the source domain is greatly reduced, resulting in the deterioration of the overall reasoning ability and the reduction of the accuracy of sketch retrieval.

Images