Deconvolution processing method, image processing method and corresponding device

Abstract

The embodiment of the invention provides a deconvolution processing method, an image processing method and corresponding devices. The method comprises the following steps: obtaining a rearrangement weight corresponding to deconvolution, wherein the rearrangement weight is obtained by rearranging an original weight of the deconvolution in an inverted order; according to the deconvolution processing parameters of the deconvolution, performing data expansion on the original input tensor of the deconvolution to obtain an expanded tensor; and according to the rearranged weight, performing convolution operation on the expanded tensor to obtain a deconvolution output tensor. According to the method, the original weight of deconvolution is preprocessed to obtain a rearranged weight, the original input tensor of deconvolution is preprocessed in a parallelizable mode to obtain an expanded tensor, deconvolution operation is converted into convolution operation with a mature parallel computing optimization scheme according to the rearranged weight and the expanded tensor, and the output tensor which is the same as the deconvolution operation result is obtained, so that the complexity of the calculation scheme is effectively reduced, the accelerated optimization is facilitated, and the consumption reduction of the deconvolution operation is realized.

Description

technical field [0001] The embodiments of the present application relate to the technical field of image processing, and specifically, the embodiments of the present application relate to a deconvolution processing method, an image processing method, and a corresponding device. Background technique [0002] Image processing mainly uses computers to analyze, process, and process graphics and image information. This technology is applied to all aspects of human life and work. In a series of transmission and processing of images, some realistic characteristics or factors will degrade the obtained images, such as inevitable noise pollution, blurred details, etc. [0003] With the continuous application and development of neural networks in the field of image processing technology, deconvolution (Deconvolution) technology is usually used to reduce or remove the problem of image quality degradation, which can map and reconstruct low-dimensional features to high-dimensional feature...

AI Technical Summary

Problems solved by technology

However, the existing deconvolution operation has high complexity and cannot achieve ideal performance optimization, which will bring huge consumption of resources and energy.

Images