Adam Świtoński

Segmentation of Fundus Eye Images Using Methods of Mathematical Morphology for Glaucoma Diagnosis

In this paper the new method for automatic segmentation of cup and optic disc in fundus eye images taken from classical fundus camera is proposed. The proposed method is fully based on techniques from mathematical morphology. Detection of cup region makes use of watershed transformation with markers imposed, while optic disk is extracted based on geodesic reconstruction by dilation. The obtained results are encouraging.

Classification of poses and movement phases

We have focused on the problem of classification of motion frames representing different poses by supervised machine learning and dimensionality reduction techniques. We have extracted motion frames from global database manually, divided them into six different classes and applied classifiers to automatic pose type detection. We have used statistical Bayes, neural network, random forest and Kernel PCA classifiers with wide range of their parameters. We have tried classification on the original data frames and additional reduced their dimensionality by PCA and Kernel PCA methods. We have obtained satisfactory results rated in best case 100 percent of classifiers efficiency.

View Independent Human Gait Recognition Using Markerless 3D Human Motion Capture

We present an algorithm for view-independent human gait recognition. The human gait recognition is achieved using data obtained by our markerless 3D motion tracking algorithm. The tensorial gait data were reduced by multilinear principal component analysis and subsequently classified. The performance of the motion tracking algorithm was evaluated using ground-truth data from MoCap. The classification accuracy was determined using video sequences with walking performers. Experiments on multiview video sequences show the promising effectiveness of the proposed algorithm.

Detection of tumor tissue based on the multispectral imaging

We have prepared multispectral image database of skin tumor diagnosis. All images have been labeled with two classes - tumor and healthy tissues. We have extracted pixel signatures with their spectral data and class assigning, thus obtained train dataset. Next we have used and evaluated the supervised learning techniques for the purpose of automatic tumor detection. We have tested Naive Bayes, KNN, Multilayer Perceptron, LibSVM, LibLinear, RBFNetwork, ConjuctiveRule, DecisionTable and PART classifiers. We have obtained results on the level of 99% classifier efficiency. We have visualized classification for example images by coloring class regions and verified if they overlap with labeled regions.

The Expanded Invasive Weed Optimization Metaheuristic for Solving Continuous and Discrete Optimization Problems

This paper introduces an expanded version of the Invasive Weed Optimization algorithm (exIWO) distinguished by the hybrid strategy of the search space exploration proposed by the authors. The algorithm is evaluated by solving three well-known optimization problems: minimization of numerical functions, feature selection, and the Mona Lisa TSP Challenge as one of the instances of the traveling salesman problem. The achieved results are compared with analogous outcomes produced by other optimization methods reported in the literature.

Feature Selection of Motion Capture Data in Gait Identification Challenge Problem

The method of discovering robust gait signatures containing strong discriminative properties is proposed. It is based on feature extraction and selection of motion capture data. Three different approaches of feature extraction applied to Euler angles and their first and second derivates are considered. The proper supervised classification is preceded by specified selection scenario. On the basis of the obtained precision of person gait identification, analyzed feature sets are assessed. To examine proposed method database containing 353 gaits of 25 different males is used. The results are satisfactory. In the best case the recognition accuracy of 97% is achieved. On the basis of classification which takes into consideration only the data of the specified segments, the ranking is constructed. It corresponds to the evaluation of individual features of the joint movements.

Spectrum evaluation on multispectral images by machine learning techniques

Multispectral pictures of skin are considered as the way of detection of regions with tumor. This article raises the problem of postprocessing of the color spectrum for the improvement of the tumor region detection accuracy. As the reference point spectra of 24 model colors were aquisited and then compared with their original spectra. Difference betweeen the original and aquisited spectra motivated the authors to use data mining nonparametrical techniques to find the measured spectra postprocessing technique. Two different approaches are described: classificational and regressional.

Kernel Postprocessing of Multispectral Images

Multispectral analysis is the one of possible ways of skin desease detection. This short paper describes the nonparametrical way of multispectral image postprocessing that improves the quality of obtained pictures. The method below may be described as the regressional approach because it uses kernel regression function estimator as its essence. The algorithm called HASKE was developed as the time series predictor. Its simplification may be used for the postprocessing of multispectral images.

Quantifying Chaotic Behavior in Treadmill Walking

The authors describe an example of application of nonlinear time series analysis directed at identifying the presence of deterministic chaos in human motion data by means of the largest Lyapunov exponent (LLE). The research aimed at determination of the influence of gait speed on the LLE value with a view to verification of the belief that slower walking leads to increased stability characterized by smaller LLE value. Analyses were focused on the time series representing hip flexion/extension angle, knee flexion/extension angle and dorsiflexion/plantarflexion dimension of the ankle. Gait sequences were recorded in the Human Motion Laboratory (HML) of the Polish-Japanese Academy of Information Technology in Bytom by means of the Vicon system. Application of the AC5000M treadmill allowed recordings in three variants: at the preferred walking speed (PWS) of each subject, at 80% of the PWS and at 120% of the PWS. According to the recommendations from the literature the LLE value was estimated twice for every time series: as the short-term LLE\(_1\) for the first stride and as the long-term LLE\(_{4-10}\) over a fixed interval between the fourth and the tenth stride. In the latter case it was confirmed that the LLE value increases with walking speed for both limbs.

Feature Extraction and HMM-Based Classification of Gait Video Sequences for the Purpose of Human Identification

The authors present results of the research on human recognition based on the video gait sequences from the CASIA Gait Database. Both linear (principal component analysis; PCA) and non-linear (isometric features mapping; Isomap and locally linear embedding; LLE) methods were applied in order to reduce data dimensionality, whereas a concept of hidden Markov model (HMM) was used for the purpose of data classification. The results of the conducted experiments formed the main subject of analysis of classification accuracy expressed by means of the Correct Classification Rate (CCR).