A novel supervised learning algorithm for salt-and-pepper noise detection
- PDF / 3,171,443 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 100 Downloads / 178 Views
ORIGINAL ARTICLE
A novel supervised learning algorithm for salt-and-pepper noise detection Yi Wang1 • Reza Adhmai1 • Jian Fu2 • Huda Al-Ghaib1
Received: 8 October 2014 / Accepted: 12 June 2015 / Published online: 23 June 2015 Springer-Verlag Berlin Heidelberg 2015
Abstract In this paper, a novel supervised learning algorithm called margin setting, is proposed to detect salt and pepper noise from digital images. The mathematical justification of margin setting is comprehensively discussed, including margin-based theory, decision boundaries, and the impact of margin on performance. Margin setting generates decision boundaries called prototypes. Prototypes classify salt noise, pepper noise, and non-noise. Thus, salt noise and pepper noise are detected and then corrected using a ranked order mean filter. The experiment was conducted on a wide range of noise densities using metrics such as peak signal-to-noise ratio (PSNR), mean square error (MSE), image enhancement factor (IEF), and structural similarity index (SSIM). Results show that margin setting yields better results than both the support vector machine and standard median filter. The superior performance of margin setting indicates it is a powerful supervised learning algorithm that outperforms the support vector machine when applied to salt and pepper noise detection. Keywords Salt and pepper noise Margin setting Noise detection Supervised learning
& Yi Wang [email protected] 1
Department of Electrical and Computer Engineering, University of Alabama in Huntsville, Huntsville, AL 35899, USA
2
Department of Electrical Engineering and Computer Science, Alabama A&M University, Normal, AL 35763, USA
1 Introduction During image acquisition and transmission via consumer electronic devices, different factors such as defects in imaging sensors and communication channels may produce additive noise [1]. Noise is an undesirable factor that distorts image features and reduces image quality. The noise removal process attempts to remove noise and retrieve image details. It is also a primary step in many image processing applications such as image compression, object recognition, and edge detection. The denoising process usually contains two-phases [3–5]. First, the noise pixels are detected. Then, a filter is utilized to estimate the original noise-free pixel values and remove noise. Ideally, noise can be removed while leaving the integrity of image details and edges preserved. There are different types of noise including impulse noise, Poisson noise, speckle noise, and Gaussian white noise. Fixed value impulse noise is also known as salt and pepper noise. It may occur due to failure in the imaging sensor cells, failure of memory locations, or synchronous errors during image transmission. Salt noise is shown as white dots with the maximum value in the image dynamic range, while pepper noise is shown as black dots with the minimum value in the image dynamic range. One of the well-known noise reduction algorithms is the standard median filter [1]. It produces satisfac
Data Loading...
