Image processing method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of image processing, in particular to an image processing method and device, computer equipment and a storage medium. The method comprises the steps of obtaining a to-be-processed target image; obtaining a neural network model, wherein the neural network model is used for discretizing the floating point type data in different data ranges by adopting different discretization modes to obtain corresponding integer data with a preset bit width; and calling the neural network model to process the target image, and outputting to obtain image feature data.According to the embodiment of the invention, the bit width of integer data is effectively controlled in the floating point type data integer process, the storage representation range of the integerdata is expanded, namely, low-bit-width and high-precision integer is realized, and meanwhile, calculation acceleration is realized due to the speed advantage and the resource advantage of fixed-pointnumber operation, so that the image processing efficiency and the subsequent image feature extraction effect are improved.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, and in particular, to an image processing method, device, computer equipment, and storage medium. Background technique [0002] Image processing algorithms based on neural networks often require deep neural network models to extract image features to complete the final tasks of classification, detection, identification, etc. In this process, large-scale vector or matrix floating point multiplication and addition operations will inevitably be introduced , considering the limitations of computing speed and computing and storage resources, it is usually necessary to accelerate the floating-point operations involved in the operation of the neural network image processing algorithm. [0003] Floating-point integerization is a means to effectively accelerate the calculation of neural network image processing algorithms and reduce the consumption of computing resources. By converting ...

AI Technical Summary

Problems solved by technology

[0004] In related technologies, the normalized nearest neighbor fixed-point number discretization method is usually used. The calculation accuracy of this method is poor, resulting in poor subsequent image feature extraction. At present, a reasonable and effective image processing method has not been provided.

Images