http://rulrepository.ru.ac.bd/handle/123456789/89 Tue, 07 Apr 2026 21:40:22 GMT 2026-04-07T21:40:22Z http://rulrepository.ru.ac.bd/handle/123456789/1113 Construction of Statistical Visual Descriptors for CBIR Exploiting Time-inhomogeneous Markov Chain Model Islam, Md. Saiful Markov stationary features (MSF) based on homogeneous Markov chain model for content-based image retrieval (CBIR) is getting popularity nowadays. It not only considers the distribution of colors that the histogram method does, but also characterizes the spatial co-occurrence of histogram patterns. However, handling a large scale database of images with large degree of heterogeneity, a simple MSF method is not sufficient to discriminate the images. The method does not capture sufficient spatial co-occurrence information as required for large databases. To overcome the shortcoming in this research, two extended methods namely multichannel nonhomogencous MSF (MCN-MSF) and multi-resolution nonhomogeneous MSF (MRN-MSF) based on original MSF are proposed. In both cases, the concept of nonhomogeneous Markov chain model is exploited to construct the features. For the first method, we incorporate spatial co-occurrence structures of different color channels of an image by applying the time inhomogeneous (nonhomogeneous) Markov chain model. For the second method, by exploiting the similar nonhomogeneous model, we incorporate the spatial cooccu1Tence information more consistently by mapping the image with different resolution. Without compromising effectiveness and robustness, the methods proposed in this paper keeps the features level simplicity. Widely recognized WANGl000 and Corell0800 databases arc used to evaluate and compare the performances of the proposed algorithms with the existing methods. The experimental results show that both methods significantly improve the performances compare to the existing methods. The results also prove that second method is more effective for large databases. This Thesis is Submitted to the Department of Computer Science and Engineering , University of Rajshahi, Rajshahi, Bangladesh for The Degree of Doctor of Philosophy (PhD) Wed, 01 Jan 2020 00:00:00 GMT http://rulrepository.ru.ac.bd/handle/123456789/1113 2020-01-01T00:00:00Z http://rulrepository.ru.ac.bd/handle/123456789/1048 Person Identification Using Gait Biometric with Challenging Clothing and Speed Cofactors Rokanujjaman, Md. Biometrics are the entire class of technologies and techniques utilized to identify the individual by using their physiological or behavioral attributes. The way of human walking, the most emergent and unique biometric signature allows automatic gait-based person identification. Gait identification task becomes more difficult due to the change of appearance by different cofactors (e.g., walking speed, shoe, surface, carrying, view, clothing and etc.). The main goal of this thesis is to develop novel methods to address the two most frequently happening covariate factors clothing and carrying condition and walking speed changes for gait recognition. These cofactors may affect some parts of gait while other parts remain unchanged and can be used for recognition. An algorithm is proposed to define which parts are more effective and which parts are less effective for cofactors like walking speed, clothing, carrying objects etc. It is found that, for clothing and carrying conditions the upper part of the body is more affected whereas for walking speed changes the lower body part is more affected. During the process of finding the effective body parts, the whole body is divided into small segments where each segment is a single row in this work. Based on positive and negative effect of each segment in terms of recognition rate, we define the whole gait into five unequal parts for clothing and carrying conditions. Usually, the dynamic areas (e.g., legs, arms swing) are comparatively less affected than static areas (e.g., head, torso) for different cofactors in appearance-based gait representation. To give more emphasis on dynamic areas and less on static areas, frequency-domain gait entropy termed as EnDFT representation is proposed and used as gait features. Experiments are conducted on two comprehensive benchmarking databases: The OU-ISIR Gait Database, the Treadmill dataset B with huge clothing variations and CASIA Gait Database, Dataset B with clothing and carrying conditions. The proposed method achieved the recognition rate 72.78% for OU-ISIR and 77.69% for CASIA gait database at rank-1 and presented better results in comparison with other existing gait recognition approaches. This Thesis is Submitted to the Department of Computer Science and Engineering, University of Rajshahi, Rajshahi, Bangladesh for The Degree of Doctor of Philosophy (PhD) Fri, 01 Jan 2021 00:00:00 GMT http://rulrepository.ru.ac.bd/handle/123456789/1048 2021-01-01T00:00:00Z http://rulrepository.ru.ac.bd/handle/123456789/546 Fingerprint Identification System Using Artificial Neural Computing Models Hossain, A.K.M. Akhtar This thesis presents the idea of features extraction technique of an off-line fingerprint. Traditionally, minutiae, core, delta, crossover, bifurcation, ridge ending, pore, island, enclosure, bridge, dot features of fingerprints are considered to identity a person. These features are available when the fingerprint images are good. The great demerit of those existing techniques is that, they fail to extract the features of poor fingerprint images. If the fingerprint image is blurred then it is difficult to extract those features from the fingerprint. In this research work is fully surveyed all the available methods and different problems related with poor fingerprint identification. Taking into consideration the merits and demerits of the available methods, a new method for feature extraction technique for the poor fingerprint images is proposed. This research is specially dealing with the problem of poorly input fingerprint identification. For these poor types of fingerprint images, features can be extracted by using proposed grid mapping feature extraction techniques. In this process, the fingerprint is processed through Grid Mapping Features (GMF) extraction with fixed square/rectangle cells, such as 16X16 square/rectangle areas. If one small square/rectangle cell of the grid map contains more than 45-55% black pixels, then its digital value is considered as I otherwise 0. These digital values are applied to the input of the neural network for training purpose using Minimum Distance Error Rate Back Propagation (MDER-BP) algorithm, which is proposed in this research work. During the training period, the values of the nodes are updated and stored in a relational knowledge base. The matching part of the system identifies the Fingerprint of a person with the help of the previous experiential values, which were stored in the relational knowledge-base of the system. The proposed Grid Mapping Feature-based fingerprint matching system gives an acceptable accuracy in off-line identification system for poor fingerprint images. It is also shown that the proposed MDER-BP Algorithm also possesses capability for training and matching the poor fingerprint images. Several factors are responsible for getting correct result through neural computing techniques. The convergence of the solution depends heavily on initialization with random numbers and accuracy of the results depends on (i) spread factors (ii) learning rates, (iii) iterations and (iv) hidden units. Finally, it has been concluded that for recognition of fingerprint using the MDER-BP Algorithm the system shows better efficiency with respect to Adaptive Resonance Theory-2 for the poor fingerprint images. This thesis is Submitted to the Department of Computer Science and Engineering, University of Rajshahi, Rajshahi, Bangladesh for The Degree of Doctor of Philosophy (PhD) Sun, 01 Jan 2006 00:00:00 GMT http://rulrepository.ru.ac.bd/handle/123456789/546 2006-01-01T00:00:00Z http://rulrepository.ru.ac.bd/handle/123456789/329 Multiple Kernel Learning And Its Application In Bioinformatics Hasan, Md. Al Mehedi During the last decades, the support vector machine (SVM) has been applied broadly within the field of computational biology or bioinformatics to answer biological questions and to reach valid biological conclusions. However, a successful application of SVM depends heavily on the determination of the right type and suitable parameter settings of kernel functions. The selection of the appropriate kernel and kernel parameters are both considered as the choice of kernel problem. Therefore, kernel learning becomes a crucial problem for all kernel-based methods like the SVM. Recently, the multiple kernel learning (MKL) has been developed to tackle the kernel learning problem efficiently and gives some scopes to improve the performance of a system. On the other hand, sometimes it is desirable to handle multiple data sources for pattern recognition in the field of bioinformatics. In this context, if these data sources are combined appropriately as one data source, it is then possible to provide a more "complete" representation of an entity which in turns, enhances the performance of a pattern recognition system. In this case, MKL also provides a way to combine features from various data sources, where each kernel will be dedicated to a particular type of data source. In order to use the above two advantages of MKL, we have applied MKL in two challenging problems in bioinformatics: protein subcellular localization prediction and protein post-translational modifications (PTMs) prediction. The knowledge of the subcellular localization and PTMs of proteins are important for both basic research and drug development. Recently various types of computational tools have been developed to predict the subcellular localization and PTMs or PTMs site of a protein through different types of machine learning algorithms. However, in order to meet the current demand of drug development and basic research, both of the above prediction systems require additional effort to produce efficient high-throughput tools. In our thesis work, we have applied MKL in order to give potential solution for the choice of kernel problem in one of the two mentioned applications of bioinformatics. In this case, the set of radial basis function (RBF) kernels (different values of sigma create different kernels) has been considered as the search space of the choice of kernel problem. Moreover, since both applications can be solved from various data sources, features from various sources are fused using multiple kernel learning with the expectation of better improvements. The experimental results show that the prediction systems using MKL based SVM provide better performance than other top existing systems in both applications. We have completed nine experiments throughout this thesis work. Where, four of those show the capability of single kernel based SVM, one shows the effects of the choice of kernel problem, one provides potential solution to the choice of kernel problem using MKL, finally, rest three show the application of MKL in handling multiple data sources. In addition to it, we have developed six user-friendly web servers for six specific prediction purposes as a product of these experiments. This thesis is Submitted to the Department of Computer Science and Engineering, University of Rajshahi, Rajshahi, Bangladesh for The Degree of Doctor of Philosophy (PhD) Sun, 01 Jan 2017 00:00:00 GMT http://rulrepository.ru.ac.bd/handle/123456789/329 2017-01-01T00:00:00Z