Image training set synthesis method and system

Abstract

The invention provides an image training set synthesis method and system. The method comprises the following steps: acquiring a background picture and a foreground picture; determining a foreground region to be added in the background picture based on a pre-trained semantic segmentation network; if it is judged that the to-be-added foreground area comprises a preset foreground adding position, adjusting the foreground picture based on a preset parameter of the preset foreground adding position to obtain an adjusted foreground picture; and based on a preset fusion algorithm, fusing the background picture and the adjusted foreground picture to obtain a new synthetic training picture. When the image training set is constructed, under the condition of less training data, a large number of needed foreground pictures are synthesized by using a small number of foreground pictures and a large number of background pictures, so that the problem of data imbalance is solved, the network training effect is improved, and the workload of artificially synthesizing the pictures is reduced.

Description

technical field [0001] The invention relates to the technical field of medical image processing, in particular to an image training set synthesis method and system. Background technique [0002] In the medical field, capsule endoscopy has the advantages of painlessness, non-injury, and large amount of image information, and has a wide range of application values. [0003] The existing technology uses artificial methods to identify and classify the original pictures taken by capsule endoscopy. In order to identify the original pictures more accurately and efficiently, it is necessary to build a model, but the model usually needs to be trained before it is used. After that, the model can be used to recognize the original image. The training data is the most important thing in the current machine learning network training, and the data volume and data quality are the key points in the network training. In practical applications, only the data collected by itself is used, and t...

AI Technical Summary

Problems solved by technology

[0005] The above two methods can indeed increase the number of pictures with a small number of categories in the training set to a certain extent, but for the first method, only sampling processing, image augmentation processing or decentralization processing is performed on the loss function. , although the number and diversity of pictures are improved to a certain extent, but due to the limited number of original pictures, an improvement space of these methods is limited to a certain extent. For scenes where the number is small or the actual scene is more diverse, This method does not help much; for the second method, use photoshop technology or similar image processing algorithms, and use manual image synthesis, which can indeed obtain more images. It can improve the network training effect to a certain extent, but because it needs to manually process the pictures, it needs to consume a lot of human resources, and it needs a lot of resource investment to get the corresponding returns, and it is difficult to use in actual production. use

Images