CNN based quick image search method

Abstract

The invention discloses a CNN (Convolution Neural Network) based quick image search method. The method includes: a first step, extracting features of an image to be searched through a CNN so as to obtain a vector feature representing the image; and a second step, performing k neighbor search on the vector feature in a feature database. The method selects CNN features based on a GOOGLENET network, which is a breakthrough in the field of computer vision after deep learning rising; the method is good in robustness; after the CNN features are extracted, based on the PQ quick search idea and an inverted strategy of text search, the method considers the personal data size during application, reasonably arranges a system parameter, and improves reordering of search results; a quick ordering strategy is adopted, and then the detection time is shortened, and the detection efficiency is improved.

Description

technical field [0001] The invention relates to the technical field of computer vision and pattern recognition, in particular to a fast image retrieval method based on CNN (convolutional neural network). Background technique [0002] In today's information and multimedia era, the Internet has entered the families of ordinary people and has become more and more important in life. In daily life, people can communicate with the outside world anytime and anywhere. In the process of our communication through the Internet, huge multimedia data is generated, but it has caused a lot of obstacles for Internet users to quickly find the information they need. Therefore, Search technology came into being. Modern people use images and videos to intuitively display their living conditions in their lives, and image information is widely used in various industries in society. Efficient retrieval and management of image information has become an urgent problem to be solved in modern societ...

AI Technical Summary

Problems solved by technology

[0010] 4) KD-TREE algorithm [4, 5] (see literature [4] Friedman J H, Bentley J L, Finkel R A. Analgorithm for finding best matches in logarithmic expected time [J]. ACM Transactions on Mathematical Software (TOMS), 1977, 3(3):209-226 and 【5】Moore A W.An intooductory tutorialon kd-trees[J].1991) idea is to divide the data into k-dimensional data space, combined with data structure retrieval algorithms such as binary retrieval Retrieval, so that building an index tree can save a lot of time compared with linear scanning, and has a good guarantee for retrieval accuracy, but when the dimension of feature data is too high, such as Gist features (see literature [3] Oliva A, Torralba A. Building the gist of a scene: The role of global image features in recognition[J].Progress in brainresearch, 2006, 155:23-36.), BOW features, the efficiency of the algorithm drops sharply

Images