Image processing method, device and mobile terminal based on convolutional neural network

Abstract

The invention discloses an image processing method, a device and a mobile terminal based on a convolutional neural network, which are suitable for the execution in a mobile terminal with a graphic program interface, wherein the convolutional neural network comprises a plurality of processing layers and a plurality of data layers and the method comprises: inputting a to-be-processed image into the convolutional neural network as first data layers; for each data layer, according to the maximum pixel value and the minimum pixel value of the data layer, converting the various pixel values of the plurality of characteristic images in the data layer into texture data; according to the sizes, the numbers of the horizontal textures and longitudinal textures of the characteristic images, combining the plurality of characteristic images after being converted into the texture data to form a corresponding big texture that is saved; for each processing layer, converting the texture data in the big texture corresponding to the previous data layer connected thereto into the first data format; and invoking the graphic program interface to render the texture data corresponding to the processing layer to generate a plurality of characteristic images for a 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

On the one hand, the data types supported by OpenGL textures are unsigned 8-bit integer (uint8), 16-bit or 32-bit floating point (float16, float32), etc., but the GPUs of mobile terminals are uneven, and only uint8 can satisfy most GPU chips for mobile terminals

On the other hand, the size of the texture supported by OpenGL is also limited. For relatively low-end GPU chips, the width and height of the texture are limited to the range of 2048px

Images