Automatic image enhancement system and method for integrating multi-scale information

Abstract

The invention discloses an automatic image enhancement system and method for integrating multi-scale information. The system comprises a sample calibration module which collects a training set image,marks a pixel level of the training set image, determines a mapping range of a label, and obtains a corresponding reference standard image, a network automatic construction module which inputs a corresponding reference standard image, specifies an image range to be perceived, automatically calculates the size and number of convolution kernels in a multi-scale fusion module, and generates a convolutional neural network model, an automatic training network module which carries out offline training of the convolutional neural network model by using a random gradient descent method optimized pixelweight loss function, and a network application network which inputs a new image into the trained network and outputs an enhanced image with the size of an original image. By integrating multi-scalefeatures, context information is effectively utilized, the image is automatically enhanced, and manual intervention is not needed.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an automatic image enhancement system and method for fusing multi-scale information. Background technique [0002] Image enhancement is one of the basic contents of image processing. According to a specific requirement, the useful information in the image is highlighted, and the useless information is removed or weakened. The purpose is to improve image quality, and the processed results are more suitable for human visual characteristics or machine recognition systems. Image enhancement technology has been widely used in medical diagnosis, aerospace, non-destructive detection, satellite image processing and other fields. [0003] Common image enhancement technologies are mostly based on image statistical information, using low-pass filtering, median filtering and other methods to remove noise in the image; using high-pass filtering, wavelet transform, etc. to enhance edges to mak...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is that the prior art cannot fuse the context information, and continuous trial and error are required

Images