Research On Pseudo-color Enhancement Algorithm Of Weld Image

For the quality of welding,it is often necessary to carry out X ray inspection of the weld and check the defects according to the weld image.This paper presents a real-time pseudo-color enhancement algorithm based on HSI color model and Retinex theory,which mainly solves the following problems:(1)In view of the serious random noise of weld image,the random noise can be smoothed effectively by selecting the appropriate frame number according to the peak signal-to-noise ratio when choosing different frame number,which is beneficial to the later pseudo-color enhancement.(2)The existing pseudo-color enhancement algorithms are mostly based on the RGB color model to fit the display equipment,but the three channels in this model have strong correlation and the color contrast of different pixels can not be enhanced according to some attribute of color.Therefore,this paper chooses the pixel self-transformation method based on the HSI color model to realize the weld image pseudo-color enhancement.However,the algorithm only applies to the images with the gray value between 0 and 255.In view of the problem that the original weld image with high dynamic range can not be directly pseudocolored,some coefficients in the original algorithm are adjusted to make the pseudo color enhancement algorithm applied to the original weld image.(3)In order to improve the contrast of defect region after pseudo-color processing,this paper combines Retinex theory and local edge preserving filter to further improve the original algorithm.The luminance channel image in the HSI color model is first input into LEP filter to obtain the base layer image representing the local mean and significant edge of the original image.The difference between the original image and the base layer image corresponding to the gray value of the pixel to obtain the detail layer image.By comparing the different parameters of the LEP filter,the obtained parameters make the detail layer retain the defect information and filter out most of the noise.Then,the output luminance channel image isacquired after the detail layer is enhanced by S function and superimpose with the base layer image,which is fused to the HSI color space and then converted to the RGB color space for display.The algorithm effectively improves the problem of low defect contrast.(4)To improve the efficiency of defect discrimination in weld seam,the proposed pseudo-color enhancement algorithm is accelerated by GPU parallel computing based on CUDA programming model,and the running time of pseudo-color algorithm is compressed effectively.The experimental results show that the image entropy and definition of the proposed algorithm are greatly improved,which can effectively improve the contrast of the image while making the defects clearer,and improve the visual resolution of the human eye to the weld image.