목표물 검출 및 인식을 위한 잡음 형태에 따른 다중 모델 가이디드 영상 필터

Abstract

In this paper, a MMGIF (Multiple Model Guided Image Filter) is proposed to eliminate the additive noise in a measured image and to eventually detect and recognize targets in the vision-used system reliably. We only consider Gaussian noise and saltand-pepper noise (impulse noise), which are practical and representative noise types in images. Therefore, the proposed MMGIF consists of two different GIFs (Guided Image Filters) and applies the appropriate GIF according to the noise type. One GIF model is a standard GIF for removing white Gaussian noise, and the other model is a Laplacian GIF for salt-and-pepper noise, which generally occurs in CCD camera images. Furthermore, in order to select the proper model, an image noise identification method is also proposed in this paper. The proposed algorithm estimates the image noise type based on a logistic regression algorithm by using kurtosis, skewness, and normality obtained from the estimated noise distribution. The performance of the proposed algorithm is evaluated in terms of peak signal-to-noise ratio and image enhancement factor through several simulations.