High-parallelism and low-delay image scaling and clipping processing method based on zynq platform

Abstract

The invention discloses a high-parallelism and low-delay image scaling and clipping processing method based on a zynq platform. The method includes the steps that original image data on the PS side ofthe zynq is obtained; a scaling algorithm including nearest neighbor interpolation, bilinear interpolation and bicubic interpolation modes and an image cropping area can be set, a corresponding transmission method is adopted on the PL side according to different interpolation modes, pixel points in each row are multiplied by corresponding coefficients of the pixel points through multi-level cache, and highly parallel pipeline is achieved. The invention can be mainly applied to the requirement of a network model for the image size in the image preprocessing stage of the neural network; the invention has certain practicability for researching the system design of the FPGA for realizing the convolutional neural network, can reduce the use of a multiplier and the problem of on-chip buffer resource shortage, and has the characteristics of real-time performance and low power consumption.

Description

technical field [0001] The invention relates to the field of image processing, and specifically refers to a zynq-based processing method of multiple interpolation mode image scaling and cutting with high parallelism and low delay. Background technique [0002] With the popularity of artificial intelligence, the application of FPGA in its field is becoming more and more extensive. Among them, the use of zynq platform to implement deep learning algorithms is now a hot topic. At present, many neural network models have been developed on FPGA-based platforms. For different network models, the size of the required input image is also different, and the quality of the image has a certain impact on the results of the entire algorithm. In addition, in the training process of the deep learning algorithm, some random cropping of the image is required to reduce the weight of the background factor and increase the stability of the model. In order to further implement the deployment of...

AI Technical Summary

Problems solved by technology

[0005] The present invention is to solve the problem that the image size in the preprocessing stage of FPGA realizes deep learning does not meet the requirements of the model, and proposes a processing method based on zynq high parallelism and low delay in multiple interpolation mode image scaling and cutting

Images