Salt-and-Pepper noise removal by median-type noise detectors and detail-preserving regularization

This paper proposes a two-phase scheme for removing salt-and-pepper impulse noise. In the first phase, an adaptive median filter is used to identify pixels which are likely to be contaminated by noise (noise candidates). In the second phase, the image is restored using a specialized regularization method that applies only to those selected noise candidates. In terms of edge preservation and noise suppression, our restored images show a significant improvement compared to those restored by using just nonlinear filters or regularization methods only. Our scheme can remove salt-and-pepper-noise with a noise level as high as 90%.

Comparative evaluation of despeckle filtering in ultrasound imaging of the carotid artery

It is well-known that speckle is a multiplicative noise that degrades the visual evaluation in ultrasound imaging. The recent advancements in ultrasound instrumentation and portable ultrasound devices necessitate the need of more robust despeckling techniques for enhanced ultrasound medical imaging for both routine clinical practice and teleconsultation. The objective of this work was to carry out a comparative evaluation of despeckle filtering based on texture analysis, image quality evaluation metrics, and visual evaluation by medical experts in the assessment of 440 (220 asymptomatic and 220 symptomatic) ultrasound images of the carotid artery bifurcation. In this paper a total of 10 despeckle filters were evaluated based on local statistics, median filtering, pixel homogeneity, geometric filtering, homomorphic filtering, anisotropic diffusion, nonlinear coherence diffusion, and wavelet filtering. The results of this study suggest that the first order statistics filter lsmv, gave the best performance, followed by the geometric filter gf4d, and the homogeneous mask area filter lsminsc. These filters improved the class separation between the asymptomatic and the symptomatic classes based on the statistics of the extracted texture features, gave only a marginal improvement in the classification success rate, and improved the visual assessment carried out by the two experts. More specifically, filters lsmv or gf4d can be used for despeckling asymptomatic images in which the expert is interested mainly in the plaque composition and texture analysis; and filters lsmv, gf4d, or lsminsc can be used for the despeckling of symptomatic images in which the expert is interested in identifying the degree of stenosis and the plaque borders. The proper selection of a despeckle filter is very important in the enhancement of ultrasonic imaging of the carotid artery. Further work is needed to evaluate at a larger scale and in clinical practice the performance of the proposed despeckle filters in the automated segmentation, texture analysis, and classification of carotid ultrasound imaging.

Novel genetic-neuro-fuzzy filter for speckle reduction from sonography images

Edge-preserving speckle noise reduction is essential to computer-aided ultrasound image processing and understanding. A new class of genetic-neuro-fuzzy filter is proposed to optimize the trade-off between speckle noise removal and edge preservation. The proposed approach combines the advantages of the fuzzy, neural, and genetic paradigms. Neuro-fuzzy approaches are very promising for nonlinear filtering of noisy images. Fuzzy reasoning embedded into the network structure aims at reducing errors while fine details are being processed. The learning method based on the real-time genetic algorithms (GAs) performs an effective training of the network from a collection of training data and yields satisfactory results after a few generations. The performance of the proposed filter has been compared with that of the commonly used median and Wiener filters in reducing speckle noises on ultrasound images. We evaluate this filter by passing the filter's output to the edge detection algorithm and observing its ability to detect edge pixels.Experimental results show that the proposed genetic-neuro-fuzzy technique is very effective in speckle noise reduction as well as detail preserving even in the presence of highly noise corrupted data, and it works significantly better than other well-known conventional methods in the literature.

An analog neural network implementation in fixed time of adjustable-order statistic filters and applications

In this paper, we show a neural network implementation in fixed time of adjustable order statistic filters, including sorting, and adaptive-order statistic filters. All these networks accept an array of N numbers Xi = S(Xi)M(Xi)2E(Xi) as input (where S(Xi) is the sign of Xi, M(Xi) is the mantissa normalized to m digits, and Ex is the exponent) and employ two kinds of neurons, the linear and the threshold-logic neurons, with only integer weights (most of the weights being just +1 or -1) and integer threshold. Therefore, this will greatly facilitate the actual hardware implementation of the proposed neural networks using currently available very large scale integration technology. An application of using minimum filter in implementing a special neural network model neural network classifier (NNC) is given. With a classification problem of l classes C1, C2,.. ., C1, NNC classifies in fixed time an unknown vector to one class using a minimum-distance classification technique.

Marginal median SOM for document organization and retrieval

The self-organizing map algorithm has been used successfully in document organization. We now propose using the same algorithm for document retrieval. Moreover, we test the performance of the self-organizing map by replacing the linear Least Mean Squares adaptation rule with the marginal median. We present two implementations of the latter variant of the self-organizing map by either quantifying the real valued feature vectors to integer valued ones or not. Experiments performed using both implementations demonstrate a superior performance against the self-organizing map based method in terms of the number of training iterations needed so that the mean square error (i.e. the average distortion) drops to the e(-1) = 36.788% of its initial value. Furthermore, the performance of a document organization and retrieval system employing the self-organizing map architecture and its variant is assessed using the average recall-precision curves evaluated on two corpora; the first comprises of manually selected web pages over the Internet having touristic content and the second one is the Reuters-21578, Distribution 1.0.

Normalizing the background and removing the trend in one-dimensional DNA fingerprint images

Maximizing an individual's genetic information from its DNA fingerprint image depends on the number of bands distinguished from the background. To approach this goal, the background should be normalized while the information is preserved. Morphological operators have been used by some authors to normalize the background for two-dimensional gel images. Methods such as mean, median and "maxpolygon" are presented in this work to normalize the background in DNA fingerprint images. Mean and median methods will lead to some deformations. Visual evaluation of the results show that the original shape of the column signals are better preserved by the maxpolygon.

An algorithm for fast adaptive image binarization with applications in radiotherapy imaging

Locally adaptive image binarization with a sliding-window threshold can be an effective tool for various image processing tasks. We have used the method for the detection of bone ridges in radiotherapy portal images. However, a straight-forward implementation of sliding-window processing is too time consuming for routine use. Therefore, we have developed a new thresholding criterion suitable for incremental update within the sliding window, and we show that our algorithm gives better results on difficult portal images than various publicly available adaptive thresholding routines. For small windows, the routine is also faster than an adaptive implementation of the Otsu algorithm that uses interpolation between fixed tiles, and the resulting images are equally good.

A successive cancellation algorithm for fetal heart-rate estimation using an intrauterine ECG signal

In this paper, we present a two-stage successive cancellation (SC) algorithm that sequentially separates fetal and maternal heartbeats from an intrauterine electrocardiogram (IuECG) signal containing both fetal and maternal QRS complexes. The ECG signal is modeled as a series of fetal, maternal, and noise events. Peak detection is first employed to locate the potential fetal and maternal QRS complexes, referred to as candidate events. Each stage automatically generates a template of a source from the candidate events in the initialization period, and thereafter performs classification of the remaining candidate events based on a template matching technique. The detected events of the stronger signal are subtracted from the composite ECG signal prior to initialization and classification of the weaker signal. Once the fetal and maternal complexes are successfully detected and separated, a counting mechanism is utilized to derive the corresponding heart rates. Computer simulation results on real IuECG data demonstrate the effectiveness of the SC algorithm.

Image enhancement based on signal subspace approach

This paper describes a block-by-block basis image enhancement algorithm which uses the signal subspace method to enhance images corrupted by uncorrelated additive noise. The enhancement is performed by eliminating the noise components in the noise subspace and estimating the clean image from the remaining components in the signal subspace.

Tomography of cells by confocal laser scanning microscopy and computer-assisted three-dimensional image reconstruction: localization of cathepsin B in tumor cells penetrating collagen gels in vitro

We used the nondestructive procedures of confocal laser scanning microscopy in combination with computer-assisted methods to visualize tumor cells in the process of penetrating collagen gels. Three independent sets of images were collected. The image information of all data sets was combined into one image, giving a three-dimensional (3D) impression at high light microscopic resolution and sensitivity. We collected information about the extracellular matrix using the reflection mode, the cell surface/morphology by staining with the fluorescent dye DiOC6(3), and the distribution of cathepsin B by Cy-3-labeled immunolocalization. The specific aim of our study was visualization of the spatial relationship of cell organelles as far as they contain the enzyme cathepsin B to cell morphology and motility in a 3D model of extracellular matrix. The majority of the enzyme was localized pericellularly, with no visible relationship to the direction of movement. However, substantial amounts also appeared in intramatrix pseudopodia and associated with the extracellular face of the plasma membrane, which may be indicative either of secretion and/or epicellular activity. Our approach has general applicability to study of the spatial relationships of cell compartments and their possible reorganization over time. This could open new horizons in understanding cell structure and function.

A fast algorithm for 1-norm vector median filtering

A major drawback with vector median filters is their high computational complexity. A fast algorithm is presented for the computation of the vector median operator based on 1-norm. The algorithm complexity is investigated both from a theoretical and an experimental point of view. Simulation results are shown proving the complexity reduction achieved by the novel algorithm.

Non-linear algorithms for processing biological signals

This paper illustrates different approaches to the analysis of biological signals based on non-linear methods. The performance of such approaches, despite the greater methodological and computational complexity is, in many instances, more successful compared to linear approaches, in enhancing important parameters for both physiological studies and clinical protocols. The methods introduced employ median filters for pattern recognition, adaptive segmentation, data compression, prediction and data modelling as well as multivariate estimators in data clustering through median learning vector quantizers. Another approach described uses Wiener-Volterra kernel technique to obtain a satisfactory estimation and causality test among EEG recordings. Finally, methods for the assessment of non-linear dynamic behaviour are discussed and applied to the analysis of heart rate variability signal. In this way invariant parameters are studied which describe non-linear phenomena in the modelling of the physiological systems under investigation.

An expert system for the detection of cervical cancer cells using knowledge-based image analyzer

Analyzing for abnormalities of cell images in the cervix uteri provides a basis for reducing deaths and morbidity from cervical cancer through detection of potentially cancerous cells, provision of prompt advice and opportunities for follow-up and treatments. However, cytopathology is usually based on subjective interpretation of morphological features. Arbitrary criteria have to be devised for their classifications. Subjective interpretations of such criteria are likely to result in diagnostic shifts and consequently disagreement occurs between different interpreters. This article presents a novel approach to the composition of segmentation and diagnosis processes for biomedical image analysis. A prototype expert system has been developed to provide an objective and reliable tool to gynaecologists. Special image analyzing techniques are used and a set of knowledge sources is designed. The expert system employs a robust control strategy which minimizes the amount of domain-specific control knowledge. It has been proved to work effectively in the detection of cervical cancer.

Median radial basis function neural network

Radial basis functions (RBFs) consist of a two-layer neural network, where each hidden unit implements a kernel function. Each kernel is associated with an activation region from the input space and its output is fed to an output unit. In order to find the parameters of a neural network which embeds this structure we take into consideration two different statistical approaches. The first approach uses classical estimation in the learning stage and it is based on the learning vector quantization algorithm and its second-order statistics extension. After the presentation of this approach, we introduce the median radial basis function (MRBF) algorithm based on robust estimation of the hidden unit parameters. The proposed algorithm employs the marginal median for kernel location estimation and the median of the absolute deviations for the scale parameter estimation. A histogram-based fast implementation is provided for the MRBF algorithm. The theoretical performance of the two training algorithms is comparatively evaluated when estimating the network weights. The network is applied in pattern classification problems and in optical flow segmentation.

Order statistics learning vector quantizer

We propose a novel class of learning vector quantizers (LVQs) based on multivariate data ordering principles. A special case of the novel LVQ class is the median LVQ, which uses either the marginal median or the vector median as a multivariate estimator of location. The performance of the proposed marginal median LVQ in color image quantization is demonstrated by experiments.

Electronic speckle interferometry, phase-mapping, and nondestructive testing techniques applied to real-time, thermal loading

A digital phase-mapping method has been developed for application in real-time electronic speckle interferometry studies. Its principles and application to a continuously deforming object are described. An efficient digital image-processing algorithm has been developed that permits quantitative interpretation of the resulting phase maps.

Automatic particle picking from electron micrographs

A computer program for automatic particle picking based on textural methods is proposed. The technique relies on the evaluation of certain textural parameters for data windows containing single particles, and those containing undesirable material. These parameters are manipulated by a discriminant analysis routine for determining the rules of classification between the different categories. The effectiveness of the method was demonstrated by application to electron micrographs of 70S Escherichia coli ribosomes.

Live cell image segmentation

A major requirement of an automated, real-time, computer vision-based cell tracking system is an efficient method for segmenting cell images. The usual segmentation algorithms proposed in the literature exhibit weak performance on live unstained cell images, which can be characterized as being of low contrast, intensity-variant, and unevenly illuminated. We propose a two-stage segmentation strategy which involves: 1) extracting an approximate region containing the cell and part of the background near the cell, and 2) segmenting the cell from the background within this region. The approach effectively reduces the influence of peripheral background intensities and texture on the extraction of a cell region. The experimental results show that this approach for segmenting cell images is both fast and robust.

Image restoration by adaptive-neighborhood noise subtraction

A new algorithm for image restoration in the presence of additive white Gaussian noise is presented. This algorithm is based on a new, adaptive method to estimate the additive noise. The basic idea in this technique is to identify uniform structures or objects in the image by use of an adaptive neighborhood and to estimate the noise and the signal content in these areas separately. The noise is then subtracted selectively from the seed pixel of the adaptive neighborhood, and the process is repeated at every pixel in the image. The algorithm is compared with the adaptive two-dimensional least-mean-squares and the adaptive rectangular-window least-mean-squares algorithms for noise suppression. The results from the application of these algorithms to synthesized images and natural scenes are presented along with mean-squared-error measures. The new algorithm performs better than the other two methods both in terms of visual presentation and mean-squared error.

OSNet: a neural network implementation of order statistic filters

A dedicated neural network model called OSNet which finds the kth largest element in an array of real numbers is proposed. Its overall processing time is constant irrespective of the number of elements in the array and is four times the processing time of a single neuron. Networks of this kind may be used as building blocks for hardware implementation of order statistic filters. Examples of using OSNet for implementing various order statistic filters and for sorting are shown.

An improved median filter

A modified median filtering technique offering improved smoothing performance while maintaining the edge-preserving ability of the conventional median filter is presented. It is shown how a mean filter outperforms a median filter in noise reduction except on or near the image boundaries. Along with the edge-preserving ability of the median filter, this observation forms the basis of the three-stage filtering algorithm which is described.

Matched filter identification of left-ventricular endocardial borders in transesophageal echocardiograms

The use of one-dimensional spatial matched filtering for identifying the left ventricular endocardial borders in human transesophageal echocardiograms recorded during surgery is investigated. A maximum-likelihood method was used to choose the endocardial intensity profiles centered within the ventricle. The computer-generated border points were compared to those identified by an experienced ultrasonographer. A 16-pixel step template located 63.2% of the border points within 2 mm of the manual border. Median prefiltering of the images reduced detection accuracy by 3% to 6%. No statistically significant difference in accuracy was found between longer and shorter templates or between data-derived and step templates. Compared to manual estimates, computer generated cross-sectional area determinations were correlated with a coefficient of 0.93. Matched filtering executes rapidly, does not require prefiltering, and performs as well as other reported methods in estimating ventricular area.

Automatic selection of macromolecules from electron micrographs by component labelling and symbolic processing

A new solution to the problem of extracting images of individual biological macromolecules from electron micrographs is described. There are three distinct steps in the process. The initial stage of low-level image processing consists of noise suppression and edge detection. An intermediate stage of component labelling and feature computation bridges the gap between the iconic (low-level) processing and the final phase of symbolic (high-level) processing. Simple symbolic objects (bounding boxes) are derived from the edges, and are easily represented and manipulated in the decision-making process. The efficacy of the algorithm is demonstrated using electron micrographs of ribosomes and ribosomal subunits. The hierarchical nature of the analysis embodies a reduction in the amount of data and a change in its nature. Initially, thousands of pixels of continuous gray levels must be dealt with. After component labelling, there are fewer than a hundred bounding boxes whose manipulation can easily be defined and articulated by an expert. The software package that has been written can thus serve as a basis for applying artificial intelligence methodologies to analysis of electron micrographs.