Research And Application Of Text Image Super Resolution Reconstruction Method

Super resolution refers to the process of restoring image details and information according to the existing image information of low resolution image.Because text has rich meaning and wide application,there is a super resolution network and method specially for text image recovery.It can be called text image super resolution.However,in the past,the super resolution method of text image is usually used to restore the low resolution image data set after artificial processing,and the restore effect of the actual low resolution image obtained from the camera in nature is not satisfactory.Therefore,in response to this problem,Text Zoom,a real text images dataset for super resolution,has been proposed in recent years.With the proposal of Text Zoom data set,a super resolution network framework TSRN for this data set was also proposed.In the past,many super resolution networks have a good recovery effect on artificially sampled low resolution images.Compared with these networks,TSRN has the optimal recovery effect on real world text image dataset Text Zoom.But looking at its structure,there are still many effective improvements that can be added to it.Therefore,Thesis makes many changes to the original TSRN framework,and verifies its effect on the Text Zoom dataset.The main research contents are as follows:(1)Making more information flow in the network and attaching importance to the context information of the image can improve performance.Based on this idea,the TSRN framework is improved in the aspects of dense connection,channel attention and parallel structure.After applying the improved model to the super resolution reconstruction task with Text Zoom data set,the performance of the image restoration task is generally better than that of the unimproved framework.(2)In order to make the network pay more attention to the text content and text boundary,the feature loss function is added to the TSRN framework.Also,the proportion of gradient loss function is increased.After applying the improved model to the super resolution reconstruction task with Text Zoom data set,the performance of the image restoration task is generally better than that of the unimproved framework.(3)By combining the above improved text image super resolution framework with the front end development knowledge,a web application micro service is constructed.It can be used for text image super resolution function.Its functions mainly include image cutting,super resolution and data management.