First scientific seminar Part I

Application of the median filtering for the reduction of impulsive noise in digital images – Bogdan Smołka (SUT)

Noise suppression in digital images and video sequences is one of the most important preprocessing steps. The importance of image enhancement techniques is mainly due to the miniaturization of the high-resolution image sensors, which enabled a rapid proliferation of new devices, that allow for the acquisition, processing and transmission of visual information.
Digital images are very often contaminated by various types of noise, introduced by malfunctioning sensors in the image formation pipeline, electronic instability of the image signal, faulty memory locations, aging of the photo sensors and storage material, transmission errors and electromagnetic interferences. Therefore, noise suppression is one of the most frequently performed image processing operations, as the enhancement of the quality of images or video streams degraded by noise is crucial for the success of further signal processing steps. Very often the quality of digital images is substantially decreased by impulse noise distortions and its removal or reduction is one of the most important image processing tasks.
The Vector Median Filter (VMF) is the most widely used method used for the removal of impulse noise in the contaminated color images. This filter is very efficient at replacing the impulses with pixels from the neighborhood, preserves sharp edges and linear trends, however it does not retain fine image structures, which are treated as noise and therefore, generally the VMF tends to produce blurry images. Additionally, the impulses injected into the image, influence the VMF output, which causes that many disturbed pixels are not removed and tend to form clusters of noisy pixels, especially in case of images polluted by high intensity impulsive noise. As a result of these unwanted features of the VMF, much research has been devoted to the construction of filters which can cope with impulsive noise while simultaneously preserving details and enhancing image.
In this work a novel, efficient extension of the vector median filter intended for the suppression of impulsive noise in color images is proposed. The new filter operates on the trimmed distances between color pixels belonging to the filtering window. The cumulated distances calculated for each pixel in the local window is used to perform the reduced vector ordering, which allows to find the pixel which is centrally located in the cluster of most similar samples. The introduced generalization allows to improve the effectiveness  of the standard vector median filter and can be used for more efficient restoration of color images distorted by high intensity impulsive noise. The unique property of the described filtering framework is its ability to sharpen the image edges which was quantified using
a novel image restoration measure. Additionally, the proposed vector median extension does not increase its computational intensity, which allows to use it in real time applications.
The proposed filter can be also used for the reduction of noise in gray scale images and applied for the enhancement of ultrasound images. Extensive experiments confirmed the favorable properties of the proposed filtering framework.

Noise reduction in ultrasound images – Krystyna Malik, Bernadetta Machała (SUT)

Ultrasound images are widely used in medical diagnostic. Usage of ultrasound has greatly expanded over the past couple of decades because this technique is more accessible, safe, less expensive and simpler to use than other medical imaging techniques. However, the quality of a medical ultrasonic image is often degraded by intrinsic artifact called speckle noise. It is a random granular pattern produced mainly by multiplicative noise that degrades the visual evaluation in ultrasound images. In medical ultrasound image processing, the reduction of speckle noise with preservation of edges and image details plays a crucial role for the diagnosis.
Many denoising techniques have been proposed for effective suppression of speckle noise, and current approaches fall into three categories including adaptive local filters, anisotropic diffusion filters, and wavelet filters.
In this work a new approach to the problem of noise removal in ultrasound images is presented. The proposed filtering design is a modification of the bilateral denosing scheme, which takes into account the similarity of pixels and their spatial distance.
The concept of the proposed Modified Bilateral Filter (MBF) is based on assigning the pixels from the filtering window a minimum connection cost of a digital path which connects them with the central pixel. In this way, each pixel is connected with the central pixel through a digital path with minimum cost function value. The connection cost is used to calculate a weight assigned to each pixel from the filtering window and the filter output is the weighted average of the pixels in a local neighborhood.
For the calculation of the weights we treat the image as a graph and utilize the Dijkstra algorithm for finding the optimal connections between the pixels (graph vertices), where the graph weights are simply the absolute differences between adjacent pixels intensities.
The Modified Bilateral Filter was compared with the other speckle noise reduction techniques in terms of the visual quality of the restored image and also in terms of objective quality measures. The control parameters were selected experimentally to obtain the best possible results in terms of objective quality measures. The gray test images were contaminatedby two types of multiplicative noise with intensities ranging from 0.05 to 0.5.
The results of the performed experiments indicate that very good restoration quality has been achieved for ultrasound images contaminated by speckle noise. The new filtering method yields significantly better results in comparison with other denoising methods both in terms of subjective quality and objective restoration measures. The beneficial feature of the proposed method is the removal of noise with preservation of edges and image details.

Implementation of selected colorization algorithms for biomedical images – Adam Popowicz (SUT)

Colorization is a process of adding colors to grayscale images and videos. Each scalar pixel value in a digital image is replaced by three values which represent the color. After successful colorization, the resulting image benefits significantly. Not only it is more pleasure for a viewer, but also the perception of hardly noticeable features increases what is crucial e.g. in the medical image analysis.
In the first part of the presentation I briefly describe four implemented colorization algorithms. Two of them are already published ideas, others are novel conceptions developed during the task realization. They include: the modified distance transformation, the isolines concept, the Lipschitz cover based colorization and the chrominance blending. The methods use similar parameters what makes them comparable. New algorithms are planned to be further investigated and we are going to publish the results soon.
In the second part I show the next result of performed task – the colorization software developed in Microsoft Visual Studio, C# language. It is a window form application enabling loading images, scribbling and saving colorized images in PNG format. For each method user can select a range for parameters and investigate their impact on the resulting images.
I also present very first attempts of colorizing the ultrasonographic images. With only one seed point and external boundary scribbling some of the algorithms properly mark target synovitis region. It is very encouraging fact relating to the project scientific goal.

Trimmed Non-Local Means Filter For Noise Removal In Digital Images – Krystian Radlak (SUT)

Increasing access to the digital acquisition systems causes a growing interest in achieving high quality images in miscellaneous applications. However, the color information can be significantly degraded during the acquisition and transmission process. Noise reduces the perceptual quality of the visual information and impedes diverse image processing tasks such as image segmentation, compression or classification. Therefore, denoising of the digital images is one of the most important steps in computer vision.
Synovital disorders may cause several limitations in joint function and body movement. Progression of structural bone damage and the degree of finger joint synovitis in rheumatoid arthritis can be assessed using ultrasound images. However, the sonographic images are contaminated by multiplicative noise and consequently a proper interpretation of the results and a correct diagnosis can be difficult. Therefore the image denoising and enhancement is strongly needed.
Multiplicative noise, also known as speckle noise is a signal distortion appearing due to signal multiplication by a noise process, and is quite difficult to remove. The main aim of the noise removal techniques is to recover the true signal from the corrupted image. The traditional methods of restoration of PDS images are based on Fourier transform and a  strategy aiming at converting the multiplicative noise into additive one and to suppress it in the frequency domain.
The main aim of this research is to apply the Trimmed Non-Local Means technique (TNLM) to PDS images to suppress multiplicative noise without the transformation to frequency domain. The TNLM  method is a modification of the Non-Local Means algorithm, in which the image pixels are restored by a weighted average of pixels, whose local neighborhood is similar to the local neighborhood of the pixel which is currently being processed. The results show, that this algorithm significantly improves  image quality, so that the denoised image is visually more pleasing and easier for interpretation.

Methods for improvement of image detail visibility – Artur Bal (SUT)

Visual judgement plays important role during analysis of many types of biomedical images especially in radiography, MRI, CT and ultrasonography. Image contras is one of the most important image features which influences on the ability of image analysis by an human expert. In many cases especially the local contrast–which is relating to the ability of recognition of small differences between grey levels of adjoining regions–has primary meaning.
The main aim of the presented work is the evaluation of usefulness of the usage of local contrast enhancement methods for the enhancement of ultrasonography images. The presented comparison is focused mainly on the usage of different version of histogram-based contrast enhancement method with the moving window. The standard algorithm of moving window contrast enhancement was compared with the Generalised Weighted Contrast Enhancement (GWCE) method. The GWCE method is based on the idea of using the weights for results obtained from the contrast enhancement method for pixels covered by the moving window; the result of contrast enhancement in GWCE method is calculated as
a weighted sum of contrast enhancement results obtained for each position of the moving window.
For the results evaluation both visual judgement and numerical function has been used. The presented results show that the usage of the tested contrast enhancement methods for ultrasonography images can improving visibility of important details at this kind of images.