An image processing method and mobile terminal based on convolutional neural network

Abstract

The invention discloses an image processing method based on a convolutional neural network and a mobile terminal. The convolutional neural network includes a plurality of data layers, a processing layer and a special activation layer. The method is executed in the GPU of the mobile terminal, and the GPU stores a A texture map, a plurality of first feature maps of the current data layer are stored in the texture map with a first data type, the method includes: obtaining a plurality of first feature maps of the current data layer from the texture map, and converting each first feature map Convert the data in the second data type into the second data type; the current processing layer renders multiple first feature maps of the second data type to generate multiple temporary feature maps of the next data layer; the current special activation layer converts each temporary feature map Convert the data in the next data layer into the predetermined value range of the next data layer, and generate multiple second feature maps of the next data layer; convert the data in each second feature map into the first according to the predetermined value range of the next data layer A data type and store it in the texture map.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to an image processing method based on a convolutional neural network and a mobile terminal. Background technique [0002] Convolutional Neural Network (CNN, Convolutional Neural Network) plays an important role in the field of image processing. For example, image classification, segmentation, style conversion, image quality improvement, etc., can all be implemented using CNN to achieve better performance than traditional processing methods better effect. At present, CNN-based image processing has been widely used on the PC side. However, its application on mobile terminals still has a bottleneck, especially when the CNN reaches hundreds of layers, a large number of floating-point multiplication operations and a large number of memory applications are required. , resulting in the computing efficiency of mobile terminals not keeping up with the development speed of ...

AI Technical Summary

Problems solved by technology

[0005] For this reason, the present invention provides a kind of image processing method and mobile terminal based on convolutional neural network, to solve or at least alleviate the problem existing above

Images