Image Blind Motion Deblurring Based on CNN-Transformer Hybrid Autoencoder

Abstract

An image blind motion deblurring method based on a CNN-Transformer hybrid autoencoder. The method includes two stages, namely a model training stage and a prediction stage. The model training stage includes the following steps: Step 1: Prepare a standard dataset for image deblurring ; Step 2: Preprocessing of experimental data; Step 3: Input the blurred image in the training set of the image deblurring standard data set into the hybrid auto-encoder part for recovery; Step 4: Combine the potentially clear image generated by the hybrid auto-encoder with its standard data The corresponding clear pictures of the target are input into the discriminator together, and the discriminator calculates the loss and returns the gradient; Step 5: The hybrid autoencoder receives the gradient from the discriminator to update the parameters; the model prediction stage includes a step: input the fuzzy picture into the training The output of the hybrid autoencoder is a clear image after deblurring. By using the present invention, better image deblurring effect can be obtained, and the image details can be recovered more clearly.

Description

technical field [0001] The invention belongs to the technical field of computer vision and image processing, and relates to an image blind motion deblurring method based on a CNN-Transformer hybrid self-encoder. Background technique [0002] As an important medium for transmitting information, images play an indispensable role in people's life and work. But the problem of image blur is common in life: due to camera shake, fast movement of objects or out-of-focus problems, the quality of the image will be reduced and the image will become blurred. In addition, various reasons: such as depth variation, occlusion of motion boundaries complicates blurring. Many shooting scenes are constantly changing over time and are not repeatable, and blurring problems caused by camera shake or fast movement of objects is very likely to result in an unusable image. How to effectively use these blurred images, the research on image deblurring methods is of great significance. [0003] Image...

AI Technical Summary

Problems solved by technology

However, most methods do not perform well enough in the balance between spatial texture details and high-level contextual information, and the visual effect is not good enough.

Images