Hardware implementation method for affine transformation in NV12 format

Abstract

The invention provides a hardware implementation method for affine transformation in an NV12 format. The method comprises the following steps: S1, reading an NV12 image, which is to, according to a bus interface protocol, acquire two rows of Y channel information and one row of UV channel image information; S2, inputting resampling, which is to carry out resampling on the images in the NV12 format, wherein the data format after resampling is actually YUV444, and the distribution is similar to that of data pixel points in a BGRA format; S3, obtaining position information of a source image through affine transformation processing, which is to obtain position information of target pixel points mapped into the source image through an affine transformation matrix; S4, performing interpolation calculation, which is to assume that a target pixel point is (dx, dy) and a source pixel point is (sx, sy), and realize that after re-sampling, for corresponding pixels distribution when interpolation calculation is performed, and for a Y channel, a U channel and a V channel, the information of four positions mapped from the target pixel points to the source image is (sx, sy), (sx + 1, sy), (sx, sy+ 1) and (sx + 1, sy+ 1); S5, performing resampling output on the image, which is to realize that the image output after resampling and interpolation calculation is YUV444, and the data of a UV channel needs to be compressed.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a hardware implementation method of affine transformation in NV12 format. Background technique [0002] In artificial intelligence technology, image recognition technology is a very important one. The process of image recognition can be roughly divided into information acquisition, preprocessing, feature extraction and selection, etc. The stage that this article focuses on is the preprocessing stage. In this stage, operations such as color space conversion, scaling, translation, and rotation are generally performed on the original image. With the continuous development of modern technology, there are many kinds of image storage formats. Currently, the commonly processed formats mainly include BGRA format and NV12 format. For these two formats, in the preprocessing stage, affine transformation is often used to realize image translation, rotation and scaling. Suc...

AI Technical Summary

Problems solved by technology

Under the same size, the data volume of the BGRA format is about 2.7 times that of the NV12 format. The increase of the data volume will inevitably lead to the increase of the area and affect the efficiency.

Images