Generation method of text detection model and text detection method

Abstract

Embodiments of the present application provide a method for generating a text detection model, a text detection method, a device, a storage medium, and a computing device. The method for generating the text detection model includes: constructing a first sub-model to be trained, and using a plurality of marked text images as training data to train the first sub-model; constructing a second sub-model to be trained, and performing training on the trained The image output by the first sub-model is marked with the position category of the text line as the training data, and the second sub-model is trained. The parameters of the second residual network are kept unchanged during training, and the parameters of the two cascaded fully connected layers are Tuning, stop training after reaching the stop condition; cascade the trained first sub-model and the trained second sub-model to obtain a text detection model. Using the embodiment of the present application can realize rapid detection of dense text.

Description

technical field [0001] The present application relates to the technical field of text detection, and in particular to a method for generating a text detection model, a text detection method, a device, a storage medium, and a computing device. Background technique [0002] This section is intended to provide a background or context to the implementations of the application that are recited in the claims. The descriptions herein are not admitted to be prior art by inclusion in this section. [0003] The current "text detection" technology has a wide range of applications and is a pre-step for many computer vision tasks, such as image search, text recognition, identity authentication, visual navigation, etc. The main purpose of text detection is to locate the position of text lines or characters in the image . Compared with the detection of general objects, text has the characteristics of multi-direction, irregular shape, extreme aspect ratio, font, color, background, etc., s...

AI Technical Summary

Problems solved by technology

[0004] At present, the common text detection methods roughly include two categories. One type of scheme is text detection based on sliding windows. Based on the idea of ​​general object detection, a large number of anchor boxes with different aspect ratios and different sizes are set, and the anchor boxes are used as sliding windows. Perform traversal search on the image or the feature map after the convolution operation of the image. For each searched position box, determine whether there is text in the box. The advantage of this method is that no complicated post-processing is required after the text box is determined. , the disadvantage is that the amount of calculation is large, it consumes computing resources and takes a long time; another kind of scheme is based on the method of calculating the connected domain (or based on the idea of ​​segmentation), using the full convolutional neural network model to extract image features, and performing Binarize and calculate its connected domain, and then use a specific method to determine the position of the text line according to different application scenarios (corresponding to different types of training data sets). The advantage of this method is that the calculation amount is small and the calculation speed is fast. The disadvantage is that the post-processing steps are cumbersome , involves a large number of calculations and optimizations in the post-processing stage. Whether the post-processing strategy is reasonable or not forms an important constraint on the overall performance of the algorithm, and there is a large room for optimization.

[0005] Regarding text detection, there are currently solutions such as "pixel aggregation network" and "differentiable binarization method". They have good performance in some open environment scene text detection, and the speed is very fast. In some specific dense text scenarios, the processing speed is greatly affected by the number of text boxes, and almost linearly decreases with the increase of the number of text boxes, which is difficult to meet the speed requirements of actual application scenarios.

Images