Convolutional neural network-based image processing method and device and mobile terminal

Abstract

The present invention discloses a convolutional neural network-based image processing method and device and a mobile terminal. The method is suitable for being executed in the mobile terminal having a graphic program interface, wherein a convolutional neural network comprises a plurality of processing layers and a plurality of data layers, and the storage parameters of the data layers corresponding to the graphic program interface are stored in the mobile terminal and comprise the characteristic graph sizes, the texture transverse number and the texture longitudinal number. The method comprises the steps of taking a to-be-processed picture as the first data layer to input in the convolutional neural network; for each data layer, and according to the storage parameter, combining the plurality of characteristic graphs of the data layer into the corresponding large texture to store; for each processing layer, obtaining the large texture corresponding the previous data layer connected with the processing layer, calling the graphic program interface to carry out the rendering processing corresponding to the processing layer to generate the plurality of characteristic graphs of the next data layer.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to an image processing method, device and mobile terminal based on a convolutional neural network. Background technique [0002] With the rapid development of convolutional neural network (CNN: Convolutional Neural Network), more and more image processing methods, such as classification, segmentation, style conversion, image quality improvement, etc., use CNN for training and learning, in order to achieve better than traditional The processing method has better effect. However, there are still bottlenecks in the application of CNN on mobile terminals, especially when CNN reaches hundreds of layers, a large number of floating-point number multiplication operations and a large amount of CPU memory application are required, resulting in the calculation efficiency and memory of mobile terminals not keeping up with CNN. speed of development. [0003] Existing methods f...

AI Technical Summary

Problems solved by technology

[0004] However, when storing data through textures, it will inevitably lead to the creation of a large number of textures, and OpenGL ES needs to continuously bind and unbind different textures during the texture rendering process, resulting in a decrease in computing efficiency

Moreover, there are a large number of convolutional layers in CNN. When using OpenGL ES to execute script rendering to achieve convolution processing, the judgment of the boundary of the feature map needs to use conditional judgment statements such as "if...else...", and such statements Easy to bring inefficiency to the rendering process

Images