Image enhancement method based on convolutional auto-encoder

Abstract

An image enhancement method based on a convolutional auto-encoder comprises the steps of processing an original image into a low-light image, and when the image is preprocessed into the low-light image, adding noise to carry out network brightening and denoising training; carrying out convolution encoding operation on the low-light image to obtain a compression feature map, and carrying out pooling operation; carrying out deconvolution operation on the obtained compressed feature map, and decoding to obtain a reconstructed bright image; calculating a loss function of the reconstructed bright image and the original image; utilizing the calculated loss function to perform parameter adjustment operation of the convolutional network, and updating parameters to optimize the network; and judgingwhether a preset number of iterations is reached, and if not, continuing the operation until the number of iterations is reached so as to complete network training. The method has the advantages thatcompared with a traditional auto-encoder image enhancement method, convolution operation is applied to the encoding stage, a large number of redundant parameters generated in the network are reduced,a better image enhancement effect is achieved, and the method can be applied to the field of image enhancement.

Description

technical field [0001] The invention relates to the technical field of image coding, in particular to an image enhancement method based on a convolutional self-encoder. [0002] technical background [0003] In modern life, high-tech photography equipment has come a long way. However, under harsh conditions, such as shooting in a low-light environment, it will lead to low visibility and low-quality defects in the image, which increases the difficulty of subsequent image processing such as image recognition and classification. For industrial production, medical research and other fields that require high image quality, low-quality images can easily make decision makers make wrong decisions. Therefore, image enhancement is crucial for image analysis applications. The image enhancement algorithm can improve the image contrast and make the enhanced picture clearer, but the contrast enhancement algorithm may appear over-enhancement or unnatural effect when processing the image. ...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that a large number of redundant parameters will be generated if the traditional encoding and decoding methods are used for three-channel color images in network training, the present invention improves the original The algorithm needs a lot of preparatory work in the early stage of network training, reduces the time of data preprocessing, and provides an image enhancement method based on convolutional autoencoder

Images