CSI Communications | January 2014 | 27 Research Front De-Noising Image Filters for Bio-Medical Image Processing Raka Kundu* and Amlan Chakrabarti** *Research Fellow, A.K. Choudhury School of Information Technology, University of Calcutta, Kolkata, India **Associate Professor, A.K. Choudhury School of Information Technology, University of Calcutta, Kolkata, India Biomedical Image Processing Biomedical Image Processing is one of the emerging domains of research work with wide and intense application in medical field. It creates a bridge between engineering and medical disciplines with a motivation to provide an overall enhancement of health care. Image processing over medical image is known as Biomedical Image Processing. The image is to be in digital form for getting processed by digital computer. Digital image is composed of pixels where every pixel has some unique location and an intensity value. The processing of this digital image by digital computer by various image processing techniques produces output either as display or generates some numeric result. Application Whenever patients go to doctors, the doctors many times prescribe medical imaging to have a good realization about the patient’s abnormality. X-ray, CT, MR and ultrasound are some common form of medical imaging. Medical imaging is the non-invasive visualization of internal organs, tissues, etc and processing of these images digitally helps us in gathering better information about the diseases related to any human organ. This information is used by doctors for better comprehension and diagnosis of diseases. In other words we can say Biomedical Image Processing makes the treatment procedure reliable, fast, cost effective and hassle free. It helps in automated identification of abnormality of organs and promotes automated medical measurement. Patients suffering from particular disease can determine the stage of the disease. It not only helps in monitoring the patient but also helps in taking precaution about any disease. These computerized developed techniques can be used at hospitals and even at remote areas where there is lack of enough number of doctors, trained technician and medical facility. To name a few, the challenging researches’ in this inter-disciplinary area are 3D modeling of human organs, cancer detection, fracture detection, infected tissue identification from microscopic image, etc. Image Noise and De-noising We must have heard about image noise. Noise is unwanted information and image noise causes occurrence of false intensity (color) value. Medical images like other images are sensitive to noise, which may get contaminated at the time of image acquisition or there may exist some other reason. Every single image detail is very vital in medical image. A small degradation or change in image intensity information may affect the post image processing steps and may lead to erroneous results. Image noise also creates disturbance in having a clear visual verification. This increases the necessity of pre-processing the image by a suitable digital de-noising technique. De-noising of image is one form of image enhancement. De-noising estimates true intensity value of pixel using other pixel’s intensity values. It enables to have a better image quality. To name a few, the recent states of the art techniques in medical image denoising are non-local means, bilateral filter, trilateral filter and anisotropic diffusion. The good noise removal results has attracted the attention of many researchers to further improve the performance of these filters in terms of de-noising quality, reduction of run time and automated parameter tuning. We have performed a comparative study of the de-noising techniques and have found non-local mean’s performance quite impressive for medical images. So we took interest to further enrich its performance. Some New Approaches The theory for non-local means image de-noising is based on self similarity of texture of the image. It gives more weight to pixels with similar neighborhoods and less weight to pixels with dissimilar neighborhood. Fig. 1 explains the theory. Each restored pixel value p in the image is the weighted average of all other pixels (q) of the image. But, chances of similar neighborhood of p and q lie in the local neighborhood around the current position p. A local neighborhood region (SxS) is defined to reduce the search space from whole image to local region. So, for non-local means each restored pixel value p in the image is the weighted average of all other pixels (q) of the confined local neighborhood region (SxS). The equations for NLM are as follows: NLMv( p) = w( p, q)v(q) where, v(p) is the restored value, w(p,q) is the weight between p and q, v(q) are pixels within the local neighborhood of current position p, h is the de-noising control parameter and d(p,q) is the Euclidean difference. Noise is an outlier whose value is quite different from the rest of the intensity population. We need to use statistical methods to reduce noise. The groups of pixels that are used for computing the new pixel value are initially arranged in ascending order. Now we know that image noise is outlier. So it is expected that outliers will reside either at the beginning or at the end of the arranged data intensity set. From Fig. 2 it can easily be said that if we are capable of discarding these outliers by mathematical techniques then we can enrich the performance of estimation of new de- noised intensity value. So we can take the help of few statistical techniques and find whether they were fruitful in enhancing the de-noising performance. One of those statistical techniques is based on alpha trimmed mean. From statistics it is known that trimmed- mean is relatively insensitive to outliers as compared to mean. Traditionally in standard Non-local means all the pixels of