Figure 1. Flow Chart of Fuzzy Process
Medical image processing is used for analyzing medical images, quantitatively. There exists many medical image modalities like Magnetic Resonance Images (MRI), Computed Tomography (CT), Ultrasound (US), etc. Among them, Ultrasound is a conventional device still practiced in medical field. The Ultrasonic image obtained from US devices is often degraded by the speckle noise. Speckle noise degrades the original image quality; so it needs an efficient denoising for despecklng it. Hence, the authors have found a non-linear denoising filter to remove the speckle noise effectively. In this paper, adaptive fuzzy based non- linear filter has been applied to various ultrasound images which got corrupted with the speckle noise. Experimental results are achieved by calculating the performance metrics such as Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR) and Signal-to-Noise Ratio (SNR), that shows the workability of the proposed approach. These results are also compared with other median based denoising filters by analytical proportion.
Digital Image Processing processes the image data for storage, transmission, and representation of the machine perception. Image processing is an alternative area of processing with signals and also with images. In image processing, linear and non-linear techniques are widely used, where linear techniques are inadequate with the non-linear techniques. Medical image processing is an essential part of the medical field; it employs with the raw data of medical images and makes it useful for visualizing the tissues and inner organs (Thirumaran and Shylaja, 2015). Different medical images are available such as Computed Tomography (CT) scan, Magnetic Resonance Imaging (MRI), Ultrasound (US) medical images etc. which needs different processing techniques called medical image processing. Noise is generated from digital image acquisition or failure of hardware and also from various environmental factors, which affects the images (Singh and Neeru, 2014). Speckle noise is a tangled development presentation which degrades the originality of the image with back scattered presence from microscopic reflections into a human body, and makes it difficult for the observer (Kumbhakarna et al., 2015).
A medical image affected by various noises and speckle noise becomes a crucial noise. To eliminate the noises in the medical images, a denoising method is needed. Denoising is performed by using various filtering approaches. Filtering plays a vital role in image processing. It involves estimation of the noise degraded in a signal. Filtering is a neighborhood operation in which pixel value of the output image is considered by means of any filtering algorithm. Filters have been working well with detailed preservation, and hence are well suited for image filtering.
A literature survey in image processing techniques using fuzzy based median de-noising filters are discussed.
Devarajan et al., (1990) analyzed the statiscal method that uses the matrix function, and proposed a new parameter called ‘column sum’ in thier paper.
Russo (1996) presented an introduction about the theory of fuzzy set and the membership function, and some of the fuzzy filters were also discussed.
Plataniotis et al., (1996) proposed a filter which is used for multichannel image processing. This method employed non-linear fuzzy membership function, and the raw data used were 55.1: 93-106.
Liang and Mendel (2000) proposed a new type-2 fuzzy filter. This filter is applied and implemented by the Bayesian equalizer channel to design fuzzy adaptive filters.
Alshennawy and Aly (2009) proposed a fuzzy logical reasoning method applied for edge detection of images, without calculating the threshold value.
Suganya and Umamaheswari, (2011) proposed a median based two stage filter, which performs adaptive Images from the filtering in two steps; first is the detection and second is the removal, by using the fuzzy concept.
Sinha and Agrawal, (2015) in their paper presented various denoising filters used to remove the noises, which degrades the image quality and detailed information.
Fuzzy based image processing is a pursuit to transfer human interpretation problems into system interpretation problems for obstructing from incomplete facts. Fuzzy image processing is stated as similar to computer perception mechanism. Fuzzy image processing is special in terms of its relation to other computer vision techniques. It also expresses the fine method of image processing by its fuzzy set (Haußecker and Tizhoosh, 1999). The filtering is performed by moderate pixels using neighborhood of nearest pixels, without affecting the textures and edges. In order to demonstrate edges and structures, fuzzy rule is applied for each value of the pixels in a specified direction (Ville et al., 2003). Fuzzy based median filter (Sukomal et al., 2012) is tested on a 8-bit gray scale images. The concept of fuzzy set is to demonstrate a group of gray level values, whereas the gray level value of threshold is between 0 and 100. The unwanted information is removed by the median based filter using fuzzy logic. Fuzzy membership functions are demonstrated by such ambiguous elevations, and the implications are achieved by the fuzzy rules (Palabaş and Gangal, 2012).
Fuzzy filter is a nonlinear filter which uses membership function based on the pixel values of the window, and the output is processed by those values of the membership function.
The Adaptive Fuzzy based Non-Linear Filter is based on mapping of gray levels as fuzzy plane (Rani, 2013). Fuzzy adaptive filter produces an immense variation in the input image, when its gray level value is higher, that is nearest to the median of the input image. An image is an array of row and column dimension i.e. M×N and gray level value, L. Membership function indicates the brightness level of an image for an image f(x, y).
A fuzzy set notation is denoted as,
In equation (1), x=1,2,..,M, y = 1,2,…N, I = The intensity xy th of (x, y) value, U = Membership value. Membership μxy function indicates reasonable properties like texture, edges, which are determined for the images locally.
Fuzzy based filtering mechanism is used for de-noising depending on fuzzy sets and it applies for selecting pixel gray level values for the image window. Denoising is performed by the median filter and the averaging filter associated with fuzzy set. The flow chart of fuzzy process is shown in Figure 1.
Figure 1. Flow Chart of Fuzzy Process
Step 1: Input-image.
Step 2:Calculate the neighborhood pixels’ gray level of image, (m × n) array window is sorted in increasing or decreasing order.
Step 3: Every neighborhood pixel has been allotted by the membership function.
Step 4: Recognize 2×k+1 pixels of (k/2≤ n2), sorted median gray level value and the last k into that sorted order.
Step 5: Membership function with the highest value is to be considered as the output.
The performance of various two stage median based denoising filters are compared with Fuzzy based Adaptive nonlinear filter as stated below.
The original ultrasonic fetal image, speckle noise corrupted image and the performance analysis of median based filters such as Median Filter (MF), and Adaptive Median (AM), Directional Weighted Median (DWM), Iterative Median Filter (IM) and Adaptive Fuzzy Based Non-Linear Filter (AFNLF) are shown in Figure 2 (a-g), respectively.
Figure 2. (a) Original Ultrasound Fetal Image, (b) Speckle Noise Corrupted Image, (c) Median Filter, (d) Adaptive Median Filter, (e) Directional Weighted Median Filter, (f) Iterative Median Filter, (g) Adaptive Fuzzy based Non-linear Filter
The significant results are tabulated in Tables 1-3 for the median based denoising filter for speckle noise corrupted ultrasound image in terms of MSE, PSNR, and SNR, respectively. From the tabulated results, it is found that the the value obtained for Adaptive Fuzzy Based Non-linear Filter is better compared with other four filters.
Table 1. Comparative Analysis of Various Denoising Filters for the Ultrasound Images (Measure: MSE)
Table 2. Comparative Analysis of Various Denoising Filters for the Ultrasound Images (Measure: PSNR)
Table 3. Comparative Analysis of Various Denoising Filters for the Ultrasound Images (Measure: SNR
The performance analysis of various denoising filters are analyzed and reported in Figures 3 to 5. From this analysis, it is inferred that Fuzzy Adaptive Nonlinear Filter yields better result that other existing filters.
Figure 3. Performance Analysis of Various Denoising Filter for Speckle Noise (Measure: MSE)
Figure 4. Performance Analysis of Various Denoising Filter for Speckle Noise (Measure: PSNR)
Figure 5. Performance Analysis of Various Denoising Filters for Speckle Noise (Measure: SNR)
In this paper, Adaptive Fuzzy Based Non-Linear Filter (AFNLF) is used for denoising the speckle noise presented in the ultrasound medical images. Adaptive Fuzzy Based Non-Linear Filer is a median based denoising filter which performs the filtering by using fuzzy sets and membership functions. AFNLF is compared with various median based denoising filters for speckle noise corrupted ultrasound images. The authors have achieved the best results for removing speckle noise AFNLF. Therefore, Adaptive Fuzzy Based Non-Linear Filter is best suited for filtering the speckle noise. They have applied this for fetal images. It can be applied to mammogram images, as well because mammogram images are highly affected with impulse noise density. For acheiving this, the algorithm can be extended to work with noises like Gaussian noise, Addictive noise, Impulse noise, Raician noise, Poisson noise etc.