Mariusz Frackiewicz

KM and KHM Clustering Techniques for Colour Image Quantisation

This paper deals with the comparison of two clustering techniques kmeans (KM) and k-harmonic means (KHM) in the case of their use in colour image quantisation. The classical KMtechnique establishes good background for this comparison. Authors proposed two original heuristic initialisation methods, one arbitrary(DC) and one adaptive (SD), that were used in both techniques. Apart from specific validity indices for clustering, the results were also evaluated by means of average colour differences in RGB (PSNR) and CIELAB colour space (ΔE) and additionally difference of colourfulness (ΔM). Experimental tests realised on benchmark colour images show the superiority of KHMover KM. Other problems with both clustering techniques e.g. empty clusters have also been highlighted.

Clustering with K-Harmonic Means Applied to Colour Image Quantization

The main goal of colour quantization methods is a colour reduction with minimum colour error. In this paper were investigated six following colour quantization techniques: the classical median cut, improved median cut, clustering k-means technique in two colour versions (RGB, CIELAB) and also two versions of relative novel technique named k-harmonic means. The comparison presented here was based on testing of ten natural colour images for quantization into 16, 64 and 256 colours. In evaluation process two criteria were used: the mean squared quantization error (MSE) and the average error in the CIELAB colour space (DeltaE). During tests the efficiency of k-harmonic means applied to colour quantization has been proved.

New image quality metric used for the assessment of color quantization algorithms

Color quantization is an important operation in the field of color image processing. In this paper, we consider a usefulness of the new DSCSI metric to assessment of quantized images. This metric is shown in the background of other useful image quality metrics to evaluate the color image differences and it has also been proven that DSCSI metric achieves the highest correlation coefficients with MOS. For further veriffcation DSCSI metric the combined methods that use to initialize the results of well-known splitting algorithms such as POP, MC, Wu etc. were tested. Experimental results of such combined methods indicate that the Wu+KM approach leading to the best quantized images in the sense of DSCSI metric.

Toward a perceptual image quality assessment of color quantized images

Color image quantization is an important operation in the field of color image processing. In this paper, we consider new perceptual image quality metrics for assessment of quantized images. These types of metrics, e.g. DSCSI, MDSIs, MDSIm and HPSI achieve the highest correlation coefficients with MOS during tests on the six publicly available image databases. Research was limited to images distorted by two types of compression: JPG and JPG2K. Statistical analysis of correlation coefficients based on the Friedman test and post-hoc procedures showed that the differences between the four new perceptual metrics are not statistically significant.

Further applications of the DSCSI metric for evaluating color quantization

Color image quantization is an often used in such tasks as image compression and image segmentation. In the paper, we continue to consider the usefulness of the new DSCSI metric for evaluating quantized images. Our use of the DSCSI metric confirmed that the color quantization in the CIELAB color space is better than in the basic RGB color space. On several examples we found very good DSCSI suitability in the case of quantization with dithering. During the tests of different dithering algorithms the best results, in terms of DSCSI metric, reached the classical Floyd-Steinberg algorithm at error propagation level of 75-85%.

A Simple Hair Removal Algorithm from Dermoscopic Images

The main goal of our work is initial preprocessing of dermoscopic images by hair removal. Dermoscopy is a basic technique in skin melanoma diagnostics. One of the main problems in dermoscopy images analysis are hairs objects in the image. Hairs partially shade the main region of interest that’s why it needs special treatment. We have developed a simple and fast hair removal algorithm based on basic image processing algorithms. The algorithm was tested on available online test database PH2 [6]. Primary results of proposed algorithm show that even if hair contamination in the image is significant algorithm can find those objects. There is still a place for improvements as long as some air bubbles are marked as a region of interest.

Colour Image Quantisation using KM and KHM Clustering Techniques with Outlier-Based Initialisation

This chapter deals with some problems of using clustering techniques K-means (KM) and K-harmonic means (KHM) in colour image quantisation. A lot of attention has been paid to initialisation procedures, because they strongly affect the results of the quantisation. Classical versions of KM and KHM start with randomly selected centres. Authors are more interested in using deterministic initialisations based on the distribution of image pixels in the colour space. In addition to two previously proposed initialisations (DC and SD), here is considered a new outlier-based initialisation. It is based on the modified Mirkin’s algorithm (MM) and places the cluster centres in peripheral (outlier) colours of pixels cloud. New approach takes into account small clusters, sometimes representing colours important for proper perception of quantised image. Pixel clustering was created in the RGB, YCbCr and CIELAB colour spaces. Finally, resulting quantised images were evaluated by means of average colour differences in RGB (PSNR) and CIELAB (\( \Delta E\)) colour spaces and additionally by the loss of colourfulness (\(\Delta M\)).

Colour Quantisation as a Preprocessing Step for Image Segmentation

Colour quantisation is very often used as an auxiliary operation in colour image processing, e.g. this operation can reduce the complexity of image segmentation process. In this chapter the results of segmentation preceded by a colour quantisation have been compared with segmentation without such preprocessing step. The choice of tools for the experiment was, for obvious reasons, limited to some colour quantisation and image segmentation methods. The colour quantisation techniques based on clustering of pixels, i.e. the classic \(k\)-\(means\) technique (KM) and new \(k\)-\(harmonic means\) technique (KHM) were considered. For image segmentation the unseeded region growing (USRG) technique has been selected from a variety of known techniques. Evaluation of the results was based on empirically defined quality function used for segmentation results. Not every method of colour quantisation, carried out as preprocessing step in the process of segmentation, leads to improved segmentation result. Therefore, our approach needs a good quantisation technique, e.g. researched segmentation technique works better for KHM quantisation technique than KM technique. This study uses different images acquired from relatively simple scenes without significant highlights and shadows. An interesting open question is what kind of colour images needs to be quantised before the segmentation. Perhaps an estimation of image segmentation difficulty will help to answer this question. The further research should be focused on establishing the conditions and parameters of additional improvement in image segmentation preceded by a colour quantisation.

The evaluation of correction algorithms of intensity nonuniformity in breast MRI images: a phantom study

The aim of this work was to test the most popular and essential algorithms of the intensity nonuniformity correction of the breast MRI imaging. In this type of MRI imaging, especially in the proximity of the coil, the signal is strong but also can produce some inhomogeneities. Evaluated methods of signal correction were: N3, N3FCM, N4, Nonparametric, and SPM. For testing purposes, a uniform phantom object was used to obtain test images using breast imaging MRI coil. To quantify the results, two measures were used: integral uniformity and standard deviation. For each algorithm minimum, average and maximum values of both evaluation factors have been calculated using the binary mask created for the phantom. In the result, two methods obtained the lowest values in these measures: N3FCM and N4, however, for the second method visually phantom was the most uniform after correction.

K-Means Color Image Quantization with Deterministic Initialization: New Image Quality Metrics.

Color image quantization is used in several tasks of color image processing as an image segmentation, image compression, image watermarking, etc. In this paper we consider four traditional (MSE, PSNR, DE76 and DM) and four new perceptual metrics (DSCSI, HPSI, MDSIs and MDSIm) as useful tools for evaluating quantized images. The values of these metrics confirm that Wu’s algorithm can be used as effective deterministic initialization of K-Means method. No empty clusters are produced by this method of quantization. The experiments were realized using 24 benchmark color images for different numbers of quantization levels. The same quantization with additional Floyd-Steinberg dithering generates the images with even better values of tested perceptual metrics.