Fuzzy density weight-based support vector scene image denoising algorithm

Abstract

The invention discloses a service robot-based scene image denoising method FDW-SVR (Fuzzy Density Weight-Based Support Vector Regression), which belongs to the field of image processing. A standard LS-SVR algorithm uses the sum of error quadratic terms to control empirical risk, so that the fitting accuracy of a model for a high distribution density sample is high, while the fitting error is large for a sparse sample polluted by noises. The influence of the uncertainties of the neighborhood information of an image and the sample density on the model is considered, and a weighting factor based on the sample distribution density is designed by using fuzzy reasoning. The fuzzy support vector machine regression model is constructed and is used for scene image denoising. In a design process of the method, a novel fuzzy membership degree design method based on the center pixel density and the neighborhood density is provided. An experimental result shows that the denoising method has better performances than the latest denoising technologies at the present in the aspects of subjective and objective evaluations.

Description

technical field [0001] The invention belongs to the field of image processing, and specifically relates to an algorithm for denoising scene images by using weighting ideas, using fuzzy reasoning to design weighting factors based on central pixel density and neighborhood density, and constructing a support vector machine regression model. Background technique [0002] In machine vision, due to various conditions and random interference, the image acquired by the imaging system usually contains a lot of noise, such as salt and pepper noise, Gaussian noise, etc., removing the noise in the image, retaining useful information, and correct understanding of the image very important. [0003] Over the past three decades, a series of denoising methods have emerged, such as traditional median filtering and Gaussian filtering. Related to the Gaussian filtering method is the Gaussian-Laplace transform method. In order to obtain better image edges, this method first performs Gaussian sm...

AI Technical Summary

Problems solved by technology

However, the existing support vector machine-based denoising algorithm does not consider the influence of sample distribution density on the regression model, and its denoising effect is not good when the noise distribution density is large or the noise distribution density is far away from the training regression model. too ideal

Images