Vol 28, No 1 (2018)

A probabilistic model of clustering ensemble based on a collection of fuzzy clustering algorithms and a weighted co-association matrix is proposed. An expression for the upper bound of the misclassification probability of an arbitrary pair of objects is obtained depending on the characteristics of the ensemble. This expression is used to determine the optimal weights of the algorithms.

We consider the strongly NP-hard problem of partitioning a finite set of Euclidean points into two clusters so as to minimize the sum (over both clusters) of the weighted sums of the squared intra-cluster distances from the elements of the cluster to its center. The weights of the sums are equal to the cardinalities of the clusters. The center of one of the clusters is given as input, while the center of the other cluster is unknown and is determined as the mean value over all points in this cluster, i.e., as the geometric center (centroid). The version of the problem with constrained cardinalities of the clusters is analyzed. We construct an approximation algorithm for the problem and show that it is a fully polynomial-time approximation scheme (FPTAS) if the space dimension is bounded by a constant.

Thinning is an important process in several applications of computer vision. It aims to find the onepixel midline of the pattern in binary image. In spite of different thinning methods that have been proposed, the existing methods are not capable of solving all thinning problems. In this work, a new iterative thinning method for binary images was proposed based on a hybrid technique of sequential and parallel approaches. It consists of three stages. The first pre-processing stage determined and prepared the contour. Next, the peeling stage tested and removed unwanted pixels. When the first two stages did not meet more pixels, the postprocessing stage prepared the skeleton to produce the final one-pixel width skeleton. In this work, the first and last stages adopted the sequential approach, while the second stage was based on the parallel approach. To evaluate the performance of the proposed method, a selected and DIBCO2010-H_DIBCO2010_GT benchmark datasets were used with benchmark measurement techniques for thinning processing. The results were compared with Huang, Zhang, K3M, and Abu-Ain methods. The experiments show that the proposed method is implemented well with all types of thinning problems, better than other methods. It is simple to design, its result skeleton has one-pixel width, and it preserves the topology and connectivity.

This paper proposes a simple approach based on Relative Flow Estimates (RFE) for shot cut detection. The property of Relative flow estimates can be used for abrupt cut detection and a correction mechanism for gradual camera-shot transition detection (e.g., fade-in and fade-out, dissolves, wipes). The exacted feature vector in each frame can be mapped into a 3-D space along the continuous time axis, and these feature data can be treated as a virtually constructed pipe with fluid flowing in the 3-D axis. Compared with existing approaches, the new RFE-based algorithm can directly detect shot cut. A wide range of test videos are used to evaluate the performance of the proposed method. The experimental results show that the new scheme can produce promising results.

This paper describes a new algorithm for adaptive selection of DCT quantization parameters in the JPEG compressor. The quantization parameters are selected by classification of blocks based on the composition of fine details whose contrast exceeds the threshold visual sensitivity. Fine details are identified by an original search and recognition algorithm in the N-CIELAB normalized color space, which allows us to take visual contrast sensitivity into account. A distortion assessment metric and an optimization criterion for quantization of classified blocks to a high visual quality are proposed. A comparative analysis of test images in terms of compression parameters and quality degradation is presented. The new algorithm is experimentally shown to improve the compression of photorealistic images by 30% on average while preserving their high visual quality.

When images are rotated and the scale varies or there are similar objects in the images, wrong matching points appear easily in the scale invariant feature transform (SIFT). To address the problem, this paper proposes a SIFT wrong matching points elimination algorithm. The voting mechanism of Generalized Hough Transform (GHT) is introduced to find the rotation and scaling of the image and locate where the template image appears in the scene in order to completely reject unmatched points. Through a discovery that the neighborhood diameter ratio and direction angle difference of correct matching pairs have a quantitative relationship with the image’s rotation and scaling information, we further remove the mismatching points accurately. In order to improve image matching efficiency, a method for finding the optimal scaling level is proposed. A scaling multiple is obtained through training of sample images and applied to all images to be matched. The experimental results demonstrate that the proposed algorithm can eliminate wrong matching points more effectively than the other three commonly used methods. The image matching tests have been conducted on images from the Inria BelgaLogos database. Performance evaluation results show that the proposed method has a higher correct matching rate and higher matching efficiency.

To measure and analyze the repeatability of the high-precision positioning control system, an optical technique based on Digital Image Correlation (DIC) method coupling with digital microscope is proposed. In this study, the reliability and feasibility of this method is verified in high-accuracy motorized linear platform, the repeatability of which is calibrated as ±0.3 μm. To determine the in-plane displacement, the DIC requires two speckles images during each hysteresis process to represent the displacement change of platform using digital microscope. In essence, the method can be considered as a “plane to plane” measurement, which can effectively reduce the random error compared to “point to point” measurements. In last, five groups of displacement data under different loadings were obtained and be used to testify the feasibility of this method. In summary, the present study provides a simple, low-cost and efficient non-contact optical technique to detect sub-micro repeatability.

Positron Emission Tomography (PET) imaging is increasingly used in radiotherapy environment as well as for staging and assessing treatment response. The ability to classify PET tissues, as normal versus abnormal tissues, is crucial for medical analysis and interpretation. For this reason, a system for classifying PET area is implemented and validated. The proposed classification is carried out using k-nearest neighbor (KNN) method with the stratified K-Fold Cross-Validation strategy to enhance the classifier reliability. A dataset of eighty oncological patients are collected for system training and validation. For every patient, lesion (abnormal tissue) and background (normal tissue around the lesion) are contoured on PET images using a semi-automatic method. Then 160 vectors are obtained to train and validate the KNN. Each vector is composed by thirty Standardized Uptake Values (SUVs) characterizing the area under investigation (lesion or background). In one case, vectors are labeled as normal or abnormal tissues by a nuclear medicine physician using a semi-automatic method; in other cases, Fuzzy C-means (FCM) and k-means are used for labelling vectors in an unsupervised manner. This study aims to evaluate the performance of the proposed classifier comparing it to the Linear Kernel Support Vector Machine (KSVM). The method accuracy is evaluated by comparison with the gold standard in terms of correct classification. Experimental results show that the KNN method achieves the highest classification accuracy using the semi-automatic labelling training (Sensitivity: 86.25%; Specificity: 90.00%; Negative Predictive Value: 88.37%; Precision: 89.81%; Accuracy: 88.12%; Error: 11.87%). In addition, the proposed method shows real-time performance; it could be applied to the field classification of PET images assisting physicians into discrimination of normal and abnormal tissue during radiation treatment planning.

Empirical mode decomposition (EMD) is an adaptive, data-driven technique for processing and analyzing various types of non-stationary signals. EMD is a powerful and effective tool for signal preprocessing (denoising, detrending, regularity estimation) and time-frequency analysis. This paper discusses pattern discovery in signals via EMD. New approaches to this problem are introduced, which involve well-known information criteria along with some other proposed ones, which have been investigated and developed for our particular tasks. In addition, the methods expounded in the paper may be considered as a way of denoising and coping with the redundancy problem of EMD. A general classification of intrinsic mode functions (IMFs, empirical modes) in accordance with their physical interpretation is offered and an attempt is made to perform classification on the basis of the regression theory, special classification statistics and some cluster- analysis algorithm. The main advantage of the innovations is their capability of working automatically. Simulation studies have been undertaken on multiharmonic signals. We also cover some aspects of hardware implementation of EMD.

This paper proposes a vehicle logo recognition (VLR) system centered on front-view cars, which has been largely neglected by vision community in comparison to other object recognition tasks. The study focuses on local features that describe structural characteristics by locating the logo of a car using a coarse-to-fine (CTF) strategy that first detects the bounding box of a car then the grille and at last, the logo. The detected logo is then used to recognize the make of a car in a reduced time. Our system starts to progress in detecting the bounding box of a car by means of a vocabulary voting and scale-adaptive mean-shift searching strategy. The system continues to process in locating the bounding box of an air-intake grille using a scale-adaptive sliding window searching technique. In the next level, the bounding box of a logo is located by means of cascaded classifiers and circular region detection techniques. The classification of vehicle logos is carried out on the patch-level as occurrences of similar visual words from a visual vocabulary, instead of representing the patchbased descriptors as bag-of-features and classifying them using a standard classifier. The proposed system was tested on 25 distinctive elliptical shapes of vehicle logos with 10 images per class. The system offers the advantage of accurate logo recognition of 86.3% in the presence of significant background clutter. The proposed scheme could be independently used for part recognition of grille detection and logo detection.

Inspired by the total margin algorithm, we extend balanced relative margin machine (BRMM) by introducing surplus variables, and propose a total margin based balanced relative (TM-BRMM). TMBRMM not only solves the loss of information points involved, but also addresses outliers at the outer boundaries that limit the maximum distance from points to separating hyperplane. Furthermore, by means of kernel function, it is easy to solve nonlinear separable datasets. The experiments on UCI datasets verify the feasibility and superiority of TM-BRMM.

The following affiliation, denoted “d,” should be included for the first author A. Al-Taie: ^dComputer Science Department, College of Science for Women, Baghdad University, Iraq