Random-shape scene character detection method and device of asymptotic regression boundary

Abstract

The invention discloses a random shape scene character detection method and device of an asymptotic regression boundary, and the method comprises the steps of extracting visual features of a to-be-detected image, carrying out the feature fusion of the visual features, and obtaining a feature expression; inputting the feature expression into a horizontal suggestion box generation network to generate a horizontal character candidate box; inputting the feature expression and the horizontal character candidate box into a direction suggestion box generation network to generate a direction character suggestion box; and inputting the feature expression and the direction character suggestion box into the character boundary generation network in any shape to obtain a scene character detection result. According to the invention, more accurate and smooth character boundaries can be generated through asymptotic regression, the positions of more accurate points are obtained through the geometric topological relation and the semantic relation between boundary sampling points, and the model has better generalization, more effective execution speed and higher detection capacity.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a method and a device for detecting a character in an arbitrary-shaped scene with an asymptotic regression boundary. Background technique [0002] In modern social life, images are a popular information carrier, which widely exist in cyberspace to convey rich information. As a more direct information carrier since ancient times, text also contains rich and accurate high-level semantic information. When text is carried by images, it can not only directly convey text information, but also help to understand the deep meaning of images. Therefore, how to detect and recognize the text in the image has very important application value in real life, which is mainly reflected in four aspects: (1) Deeply intelligent visual question answering or description system. For a given image, the machine can intelligently combine the text information in the image to answer or ...

AI Technical Summary

Problems solved by technology

Due to the interference of the background, many misjudgments are likely to occur, such as judging the pixels of the text area as the background, or misjudge the background as the text area

Moreover, such methods are usually difficult to produce smooth boundaries, which will have a negative impact on some practical applications

In addition, this type of method usually involves complex post-aggregation processing, which requires processing a large number of pixels, thereby slowing down the speed of the entire algorithm

The method based on the regression segmentation hybrid also involves using the region proposal network to generate text candidate boxes, and the segmentation is limited to the candidate boxes. When the candidate box is not accurately positioned, its segmentation will also be affected.

In addition, this type of method also involves segmentation of multiple scales and a large number of candidate frames, and its execution speed is also seriously affected

Images