Multi-instance natural scene text detection method based on relevance level residual errors

Abstract

The invention provides a multi-instance natural scene text detection method based on relevance level residual errors. According to the method, a feature extraction network is adopted to extract coarse-grained and fine-grained combined multi-scale features by utilizing relevance level residual errors and reverse level-by-level feature fusion, and more accurate and more complete text information iscontained, so that the text detection accuracy is improved; secondly, text detection box regression loss used in the method is composed of two parts, namely CIoU Loss and angle loss; particularly, theuse of CIoU Loss considers the factors such as the overlapping area, the center distance and the length-width ratio between the prediction text detection box and the real text detection box, so thatthe actual regression condition of the text detection box can be evaluated more accurately, and the performance of the text detection method can be improved; then, the hardware calculation pressure isreduced by adopting a proper mode in multiple steps, and finally, the detection effect on a conventional text region and a small text region is very good.

Description

technical field [0001] The invention belongs to the fields of deep learning, computer vision and text detection, and in particular relates to a multi-instance natural scene text detection method based on relevance hierarchical residuals. Background technique [0002] As a main way of information transmission, text plays an indispensable role in our daily life. With the advent of the era of big data, how to obtain text information in massive images has become an urgent problem to be solved. . Therefore, based on the development of deep learning, natural scene text detection technology has become a very popular research direction in the field of computer vision, which is of great significance for image retrieval and scene understanding. [0003] At present, the advent of a large number of research results has made natural scene text detection widely used in various industries and fields. For example, many Internet companies have developed image retrieval, street view navigat...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a multi-instance natural scene text detection method based on the relevance level residual to solve the problem that the current text detection method is not effective in detecting small text areas, and the loss function commonly used in text detection is not very good The problem of evaluating the actual regression of text detection boxes

Images