Multilayer vision expression and deep network-based hand-drawn image retrieval method

Abstract

The invention belongs to the field of computer vision and deep learning, and particularly discloses a multilayer vision expression and deep network-based hand-drawn image retrieval method. The retrieval method comprises the following steps of s1, decomposing multilayer vision expression of a hand-drawn image and a natural image; s2, designing a framework of a multilayer deep learning convolutionalnetwork, and learning multilayer semantics of the image; s3, performing feature extraction on different layers of the hand-drawn and natural images; s4, performing deep feature fusion of multilayer vision; and s5, performing similarity calculation to obtain an optimal retrieval result. The method has the beneficial effects that 1, through layering operation on the hand-drawn and natural images, semantic information, spatial information and detail features of the hand-drawn and natural images are greatly expanded; and 2, existing hand-drawn image database and natural image database can be automatically expanded, and under the background of big data, the hand-drawn image retrieval precision can be improved by 20-30%.

Description

technical field [0001] The invention belongs to the field of computer vision and deep learning, and relates to a hand-painted image retrieval method based on multi-layer visual expression and deep network. Background technique [0002] Hand-painted graffiti is a very intuitive and versatile tool for human beings. It has been used to describe the real visual world in which human beings exist since ancient times. Research has shown that hand-drawn pictures have the same mechanism as real pictures to activate the visual area of ​​the human cerebral cortex. In recent years, with the increasing number of touch-screen devices, such as touch-screen mobile phones and touch-screen tablet computers, research on hand-drawn image retrieval technology has become increasingly prosperous and important. Compared with the traditional retrieval based on text and image content (such as Google image search engine), the advantages of hand-drawn image retrieval such as flexibility, fresh shoppin...

AI Technical Summary

Problems solved by technology

[0004] However, there is a huge difference between hand-drawn sketches and natural pictures: because hand-drawn sketches exhibit a highly abstract visual expression compared to natural pictures, using existing methods to extract features from hand-drawn sketches, the resulting feature descriptors The content of hand-drawn sketches cannot be fitted effectively; for the same object, there is a huge gap in the description and expression of hand-drawn sketches by different groups of people, which makes the matching of hand-drawn sketches and natural pictures more difficult; at the same time, the hand-drawn Mapping sketches and natural images to the same visual domain is also a difficult task

Images