Hand-drawn sketch searching method based on deep learning

Abstract

The invention discloses a hand-drawn sketch searching method based on deep learning. The method comprises the steps of inputting the pictures in a picture database into an edge feature extractor to obtain the edge features, inputting the edge features into a depth feature extractor to obtain a line feature representation vector of an picture, storing the line feature representation vector through an picture feature persistence service, performing binarization compression on the edge depth feature vector of the picture to be searched, and caching the edge depth feature vector into a picture search engine; sending a hand-drawn sketch inputted by the user to the feature extractor, and obtaining a line feature representation vector; using the feature vector and the required most similar picture number to request for a picture searching engine; obtaining a feature vector by using the picture search engine, and then carrying out the binarization processing, carrying out similarity calculation on the feature and picture features in a cache, sorting the most similar picture IDs of a corresponding request number according to the similarity, and then requesting the picture database, indexing the pictures by using the IDs, and returning a result to a requester.

Description

technical field [0001] The invention relates to the technical field of graphic retrieval, in particular to a deep learning-based hand-drawn sketch image search method. Background technique [0002] Search by image is a technology that searches for similar images by inputting a picture, and provides users with a search technology for searching related graphic and image data. It involves many disciplines such as database, data cache, computer vision, image processing, and information retrieval. Its key technology lies in feature representation and similarity measurement. The image search service is widely used in various fields such as big data image retrieval, Internet image material search, and shopping search. At present, the use of classifiers based on deep neural networks can achieve very satisfactory results in vision tasks. [0003] Karen Simonyan and Andrew Zisserman proposed a deep neural network structure for classification, known as the VGG (Visual Geometry Group...

AI Technical Summary

Problems solved by technology

[0007] However, the traditional image search is based on existing images. However, there is a situation that the image requested as a search does not exist, and there are only hand-drawn sketches for conceptual expression. People hope to search through simple hand-drawn sketches. Produce images that match in form or concept

Images