Trademark detection method based on convolutional neural network

Abstract

The invention discloses a trademark detection method based on a convolutional neural network. According to the method, various kinds of trademark pictures and pictures without trademarks are collected, and the pictures with the trademarks are marked; the convolutional neural network is initialized and is trained through trademark samples and non-trademark samples. In the picture testing process, candidate windows which may contain the trademarks in pictures to be tested are selected through a target area selection method, and color space conversion and scaling processing are carried out on the candidate windows; then the candidate windows are input into the convolutional neural network for recognition, and the candidate windows which are recognized to be the trademarks are marked in the pictures to be tested. According to the method, target area characteristic extraction and recognition are combined through the convolutional neural network, uncertainty caused by characteristic design is avoided, besides, good invariability is maintained during rotation, translation and scale changing, detection speed is increased Based on selection of the divided target area, and meanwhile the false detection rate is lowered.

Description

technical field [0001] The invention belongs to the field of target detection and recognition, and relates to a method for detecting a specific target, especially a trademark, from an image. Background technique [0002] At present, with the rapid development of a large number of information media such as TV and the Internet, a large amount of advertising information is flooding people's lives. How to analyze the source of advertisements through the trademarks contained in advertisements and effectively filter the information to ensure the effectiveness of advertisements Sexuality, and the amount of information consumers receive, becomes a matter of concern. [0003] Affected by scale transformation, viewpoint transformation, lighting conditions, occlusion, background interference, etc., accurate detection and recognition of trademarks in complex scenes is a challenging task. Most of the existing target detection and recognition methods are based on traditional features. Wh...

AI Technical Summary

Problems solved by technology

Most of the existing target detection and recognition methods are based on traditional features. When the trademark is affected by angle changes, lighting conditions, and slight deformation, the recognition rate decreases and cannot be applied to daily life.

On the other hand, the traditional method uses a multi-scale sliding window method to traverse the image to find the target. This method leads to a sharp increase in the number of windows that need to be identified, and the real-time performance is relatively poor.

Images