Image processing method and device, electronic equipment and storage medium

Abstract

The invention provides an image processing method and device, electronic equipment and a storage medium. The method comprises the steps of inputting a to-be-processed image into a pre-trained convolutional network model; processing the to-be-processed image through the convolutional network model; and outputting a processing result of the to-be-processed image. The convolutional network model comprises a convolutional network module with linear phase constraint. The convolutional network module comprises a deep convolutional network layer and a linear phase point-by-point convolutional networklayer. The deep convolution network layer adopts a convolution kernel with the size of 3*3. The linear phase point-by-point convolution network layer adopts a linear phase point-by-point convolutionkernel with the size of 1*1, and the weight of the linear phase point-by-point convolution kernel is symmetric or antisymmetric in the depth direction. According to the invention, the image is processed based on the convolutional network model with the channel number reduction function, so that the parameter quantity in the image processing process can be effectively reduced, and the image processing complexity and the calculation amount of image processing equipment are reduced.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to an image processing method, device, electronic equipment and storage medium. Background technique [0002] The artificial intelligence technology represented by the convolutional neural network algorithm has achieved tremendous development in recent years and has been widely used in the field of computer vision, and has been successfully applied in image recognition, image compression, and target detection. However, there are a large number of redundant parameters in the traditional convolutional neural network structure, and the calculation complexity is too high. Faced with many difficulties. [0003] In related technologies, the following methods can be used to overcome the above-mentioned problems: [0004] (1) Optimize the trained complex neural network model by using neural network compression and acceleration methods, such as pruning, low-rank decomposit...

AI Technical Summary

Problems solved by technology

[0006] However, in the above method (1), the pruning method is very unfriendly to hardware, and it is difficult to take advantage of hardware acceleration; low-rank decomposition is already difficult to achieve in networks designed with small-scale convolutions. Compression and acceleration; quantization will have a certain loss of precision, it is difficult to take advantage of the algorithm for general-purpose hardware; knowledge distillation theory is complicated, it can only be used for classification tasks with softmax function, and the performance is often limited, still in the theoretical developing

In the above (2) method, for various lightweight convolutional neural network structures, there are certain limitations in reducing the amount of parameters and computational complexity, and there is still room for improvement

Images