Semi-supervised cross-media feature learning retrieval method

Abstract

The present invention provides a kind of cross-media characteristic learning retrieval method based on semi-supervised, comprises the following steps: Step 1: build multimedia database, step 2: seek the projection matrix of different media types; (2.1) define objective function: (2.2) target Function optimization: (2.3) Project the original features of the multimedia data to the public space, step 3: perform cross-media retrieval; (3.1) Extract the features of the media data submitted by the user: use pre-trained model to extract the features of the data, (3.2) project the feature vector of the media data into the common space (3.3) calculate the similarity between the projected feature vector and other vectors in the common space, (3.4) Return the top k media data with the highest similarity. The method of the invention reduces computational complexity, realizes robustness to noise, and improves retrieval accuracy.

Description

technical field [0001] The invention relates to a semi-supervised-based cross-media feature learning retrieval method, which belongs to the field of retrieval. Background technique [0002] With the development of modern computer science and technology, multimedia data such as images, texts, and videos on the Internet are increasing rapidly. Content-based multimedia retrieval has become more and more important, and a lot of research has been done on it. Traditional content-based retrieval methods usually focus on single mode retrieval, such as image retrieval, text retrieval. In this case, query and retrieval results are of the same media type. However, single-mode retrieval cannot fully exploit diverse media data. To solve this problem, cross-media retrieval has been proposed and becomes more and more important. It is designed to query one type of data as a query to retrieve related data objects of another type. For example, users can use pictures of tigers to retrieve...

AI Technical Summary

Problems solved by technology

However, the dimensionality of the original features is usually large, which affects the computational complexity of the algorithm

Images