Abstract: Abstract: The richness of the marbling in beef, as an important index of beef quality, can be used to characterize the beef fat content. In particular, the area ratio of marbling, big fat density, and small fat density are the main indicators for most existing beef grade determination. Researchers have investigated that computer vision and image processing is applicable to the automatic grading of beef marbling, and thus plays a great role in promoting the development of the beef industry. However, images may be polluted when experiencing acquisition, transmitting and other processing. Consequently, the quality of the images may be reduced, and thereby, more uncertainties emerge. Importantly, the texture of the beef marbling image becomes blurred and texture contour is not clear. It will further affect the subsequent procedures of texture segmentation and extraction. Therefore, it is necessary to use the de-noising method with better edge preserving property to keep the edge and texture information of the image. In this study, we aimed to use the method of multi-scale interval interpolation wavelet to de-noise images, and thereby to smooth the gray values to segment and extract the regions of beef muscle, large and small fat particles from the beef marbling image. Here, we used the method of multi-scale interval interpolation wavelet to solve the partial differential equation, thus to de-noise images. Specifically, from this method, the edge-preserving smoothing for different object area can be realized, so that the texture and edge of beef marble were made more clearly. In addition, in this method, we chose the external collocation points adaptively, thus the computational efficiency can be greatly improved. In particular, extension method based on Center Similarity Transformation can be used to solve the boundary effect effectively. Firstly, on the basis of the objective evaluation index of the image, the PSNR (Peak Signal to Noise Ratio) mean value of the image de-noised by the proposed algorithm was higherthan the mean values obtained by using the mean filtering, median filtering and Wiener filtering of 9.0, 8.2 and 6.6 dB, respectively. In addition, the SSIM (Structural Similarity Image Measurement) value of the image de-noised by the proposed algorithm was also the largest among values obtained by algorithms mentioned above. Secondly, it is known that different objectives have different gray values, which is taken as the principle of segmentation. Hence, the processed image was segmented using different gray thresholds. In detail, the procedures of segmentation included two steps. The first step was to obtain the gray thresholds by prior knowledge, and the next was to segment the image for dividing the background, external fat, adhesive fat, small and big fat from the image by those thresholds. Finally, we compared the results of segmentation derived from our methods with the results of segmentation from Otsu. Here we showed that using the de-noised method of multi-scale interval interpolation wavelet was useful to achieve a local uniform smooth and keep the object contour information of beef images, thus to improve the accuracy of the segmentation and extraction of fat particles. The result of segmentation by gray thresholds was more accurate than the results of Otsu and retained more details about the texture of beef marbling. Furthermore, we also found that our result almost had no omission in segmenting and extracting fat particles, especially for small fat particles. Overall, our results provided a new de-noised method to improve the accuracy of beef grading.