International Journal of Computer Science and Technology

International Journal of Computer Science and Technology | Vol. 13, No. 1, January 2011 | pp. 1–7
DOI: 10.46882/2011/IJCST/000107

Research Article

Title: A Scalable Content-Based Image Retrieval System Using Color Coherence Vectors and Wavelet Decomposition

Names of Authors: Li Wei¹, Francois Dubois²

Authors’ Affiliations: ¹Department of Computer Science, Tsinghua University, Beijing, China; ²Laboratoire d'Informatique, Sorbonne Université, Paris, France

Abstract: The exponential growth of multimedia repositories demands highly scalable Content-Based Image Retrieval (CBIR) methods capable of extracting relevant images with minimal computational complexity. Conventional systems based solely on global color histograms lose critical spatial data, returning high false matches. This paper presents an advanced CBIR architecture combining Discrete Wavelet Transform (DWT) and Color Coherence Vectors (CCV). The system applies a two-level Haar DWT to decouple structural textures into sub-bands, while spatial distributions of color profiles are derived via CCVs to partition pixels into coherent and incoherent clusters. Multi-dimensional indexing is optimized through an R-tree structural model to accelerate database query execution times. Experiments were performed on the Corel 1,000-image benchmark collection. The combined DWT-CCV approach yielded a mean precision rate of 84.6% and a recall rate of 79.2%, outperforming standard histogram-based approaches by over 18.5%. The query execution runtime averaged 124 milliseconds per search on a standard database setup, establishing high scalability characteristics. This framework provides an industrial-grade solution for digital asset management, stock photo licensing platforms, and online product catalog search systems.

Keywords: Content-Based Image Retrieval, Discrete Wavelet Transform, Color Coherence Vector, Feature Extraction, Computer Vision

Manuscript Timeline: Received August 20, 2010; Revised September 25, 2010; Accepted October 06, 2010; Published January 18, 2011.

Citation: Wei, L., & Dubois, F. (2011). A Scalable Content-Based Image Retrieval System Using Color Coherence Vectors and Wavelet Decomposition. International Journal of Computer Science and Technology, 13(1), 1–7.

International Journal of Computer Science and Technology | Vol. 12, No. 12, December 2010 | pp. 83–89
DOI: 10.46882/2010/IJCST/000106

Research Article

Title: Design and Implementation of an Automated Smart Home Energy Management System Using ZigBee Networks

Names of Authors: Marcus E. Vance¹, Sarah J. Connor²

Authors’ Affiliations: ¹Department of Electrical and Computer Engineering, Austin, USA; ²Smart Systems Engineering Group, Melbourne, Australia

Abstract: Rising global energy costs and carbon footprint considerations require residential infrastructure to implement smart power management schemes. This paper develops a hardware and software prototype for an Automated Smart Home Energy Management System (ASHEMS) built upon IEEE 802.15.4/ZigBee wireless mesh networks. The infrastructure links multiple end-node microcontrollers connected to domestic appliances to monitor instant voltage, current, and real power consumption values. A central home gateway running an embedded Linux kernel collects data over the air and executes a priority-based load shedding algorithm during peak pricing intervals. Experimental results across a 30-day domestic deployment showed an average reduction of 18.6% in total electrical energy consumption. Peak demand overheads dropped by 31.2% through scheduled deferment of heavy thermal loads (such as water heaters and air conditioning units). The ZigBee mesh protocol maintained a robust packet reception rate of 98.4% across a 50-meter indoor range, successfully traversing brick partitions and defeating 2.4 GHz Wi-Fi interference. The system's passive power consumption remained negligible at less than 1.2 Watts, confirming its viability as an eco-efficient deployment for residential automation.

Keywords: Smart Home, ZigBee Network, Energy Management, Embedded Systems, Load Shedding, Automation

Manuscript Timeline: Received August 11, 2010; Revised September 19, 2010; Accepted October 03, 2010; Published December 15, 2010.

Citation: Vance, E. M., & Connor, J. S. (2010). Design and Implementation of an Automated Smart Home Energy Management System Using ZigBee Networks. International Journal of Computer Science and Technology, 12(12), 83–89.

International Journal of Computer Science and Technology | Vol. 12, No. 11, November 2010 | pp. 76–82
DOI: 10.46882/2010/IJCST/000105

Research Article

Title: An Enhanced Edge Detection Algorithm for Medical Image Segmentation Based on Fuzzy Logic

Names of Authors: Amina T. Al-Farsi¹, Yuki Tanaka²

Authors’ Affiliations: ¹Biomedical Engineering Department, King Saud University, Riyadh, Saudi Arabia; ²Graduate School of Information Science, Kyoto University, Kyoto, Japan

Abstract: Accurate edge detection in medical images is a critical prerequisite for clinical diagnostics, tumor boundaries identification, and computer-aided surgeries. Classic differential operators such as Sobel and Canny frequently fail when applied to Magnetic Resonance Imaging (MRI) scans due to low contrast and high speckle noise levels. This paper introduces an Enhanced Fuzzy Edge Detection Algorithm (EFEDA) that incorporates adaptive membership functions to resolve pixel ambiguities. The proposed system analyzes a 3 * 3 pixel neighborhood window, computing intensity gradients which are subsequently mapped to fuzzy linguistic sets (Low, Medium, High) using Gaussian functions. Fuzzy rules determine whether a pixel belongs to a true boundary, background, or noise artifact. Tests performed on a standard dataset of 200 brain MRI images demonstrated that EFEDA achieves a Peak Signal-to-Noise Ratio (PSNR) improvement of 4.8 dB over the standard Canny filter. Furthermore, the Pratt's Figure of Merit scored 0.89, highlighting superior boundary preservation properties under 15% salt-and-pepper noise conditions. The visual results confirmed clearer delineation of micro-structures without introducing artificial artifacts, facilitating automated volumetric assessments of cerebral lesions.

Keywords: Image Segmentation, Fuzzy Logic, Edge Detection, Medical Imaging, MRI, Noise Reduction

Manuscript Timeline: Received July 05, 2010; Revised August 14, 2010; Accepted September 01, 2010; Published November 15, 2010.

Citation: Al-Farsi, T. A., & Tanaka, Y. (2010). An Enhanced Edge Detection Algorithm for Medical Image Segmentation Based on Fuzzy Logic. International Journal of Computer Science and Technology, 12(11), 76–82.

International Journal of Computer Science and Technology | Vol. 12, No. 10, October 2010 | pp. 69–75
DOI: 10.46882/2010/IJCST/000104

Research Article

Title: Parallel Matrix Multiplication Optimization on Multi-Core Architectures Using OpenMP

Names of Authors: David H. Green¹, Carlos M. Silva²

Authors’ Affiliations: ¹High Performance Computing Lab, Technical University, Munich, Germany; ²Department of Computer Engineering, University of Lisbon, Lisbon, Portugal

Abstract: Matrix multiplication constitutes the fundamental arithmetic kernel in digital signal processing, computer graphics, and numerical physics simulations. As physical microprocessors hit power walls, performance scaling relies heavily on multi-core exploitation and cache-coherency management. This research investigates the optimization of dense matrix-matrix multiplication (C = A * B) on standard multi-core architectures using the OpenMP API. We present a dynamic loop-tiling strategy combined with continuous memory alignment to maximize CPU Level 1 and Level 2 cache hit ratios. Computations were executed on dual-quad-core Intel Xeon processors running at 2.66 GHz. The experimental findings indicate that a tile block size of 64 * 64 minimizes cache misses by 42.1% compared to naive nested-loop approaches. The optimized parallel algorithm reached an execution speedup factor of 7.42 on an 8-core hardware setup, reflecting a parallel efficiency of 92.75%. Memory bandwidth bottlenecks were minimized using thread-affinity pinning, which prevents expensive inter-socket cache invalidations. The mathematical error verification confirmed that floating-point accuracy stayed within standard precision boundaries (epsilon <= 10^-15). This framework serves as a scalable baseline for modern high-performance linear algebra libraries.

Keywords: Parallel Computing, OpenMP, Cache Optimization, Loop Tiling, Multi-Core Processors, Matrix Multiplication

Manuscript Timeline: Received May 28, 2010; Revised July 10, 2010; Accepted August 05, 2010; Published October 20, 2010.

Citation: Green, H. D., & Silva, M. C. (2010). Parallel Matrix Multiplication Optimization on Multi-Core Architectures Using OpenMP. International Journal of Computer Science and Technology, 12(10), 69–75.

International Journal of Computer Science and Technology | Vol. 12, No. 9, September 2010 | pp. 62–68
DOI: 10.46882/2010/IJCST/000103

Review Article

Title: A Comprehensive Survey of Machine Learning Techniques for Credit Card Fraud Detection

Names of Authors: Sanjay V. Patel¹, Mei Ling Chang²

Authors’ Affiliations: ¹Department of Data Analytics, Indian Institute of Technology, Mumbai, India; ²Department of Computer Science, National University of Singapore, Singapore

Abstract: Financial institutions suffer multi-billion-dollar losses annually due to sophisticated credit card fraud operations. Machine learning algorithms have emerged as crucial barriers, yet selecting optimal configurations remains challenging given extreme data imbalance issues, where fraudulent transactions account for less than 0.1% of total logs. This paper reviews over 50 prominent literature sources published between 2005 and 2010 focusing on supervised, unsupervised, and hybrid machine learning frameworks. We evaluate algorithms including Support Vector Machines (SVM), Random Forests, Artificial Neural Networks (ANN), and Hidden Markov Models (HMM) based on their detection rates and false-positive frequencies. The survey analyzes data preprocessing approaches, specifically Synthetic Minority Over-sampling Technique (SMOTE) and principal component analysis, highlighting their role in resolving data skewness. Our comparative matrix indicates that hybrid models joining unsupervised clustering with supervised classification yield the highest Area Under the Curve (AUC) values, averaging 0.94 ± 0.02. We also identify open challenges such as real-time processing constraints, concept drift where fraud patterns evolve dynamically, and privacy concerns during cross-bank data sharing. Finally, we map future research trajectories in deep network architectures.

Keywords: Fraud Detection, Credit Card Security, Machine Learning, Data Imbalance, SMOTE, Comparative Survey

Manuscript Timeline: Received June 02, 2010; Revised July 15, 2010; Accepted August 02, 2010; Published September 15, 2010.

Citation: Patel, S. V., & Chang, M. L. (2010). A Comprehensive Survey of Machine Learning Techniques for Credit Card Fraud Detection. International Journal of Computer Science and Technology, 12(9), 62–68.

International Journal of Computer Science and Technology | Vol. 12, No. 8, August 2010 | pp. 55–61
DOI: 10.46882/2010/IJCST/000102

Research Article

Title: A Hybrid Cryptographic Framework for Securing Cloud Storage Services Using AES and RSA

Names of Authors: Robert K. Johnson¹, Elena R. Petrova²

Authors’ Affiliations: ¹School of Computing, Cyber Security Center, Boston, USA; ²Faculty of Informatics, Technical University, Sofia, Bulgaria

Abstract: Cloud storage adoption is heavily restricted by data privacy and security vulnerabilities during multi-tenant data access. Single encryption models either suffer from heavy computational bottlenecks or lack robust key management schemes. This paper presents a hybrid cryptographic framework designed to maximize data confidentiality in distributed cloud environments. Data streams are symmetrically encrypted using the Advanced Encryption Standard (AES) with a 256-bit key length to ensure high throughput. Concurrently, the AES secret keys are asynchronously protected using Rivest-Shamir-Adleman (RSA) 2048-bit encryption during distribution phases. The proposed framework implements a zero-knowledge proof scheme ensuring that the cloud service provider retains no visibility into the plaintext files or encryption keys. Experimental evaluations showed an encryption throughput of 45 MB/s for large-scale datasets, outperforming stand-alone asymmetric models by a factor of 10. The processing overhead remained below 8% across standard virtualized cloud instances. Security analysis under simulated brute-force and side-channel threats verified that the hybrid approach provides exponential computational resistance against key leakage. This model addresses enterprise privacy compliance needs without sacrificing file transfer performance.

Keywords: Cloud Security, Advanced Encryption Standard, RSA Encryption, Hybrid Cryptography, Data Confidentiality

Manuscript Timeline: Received April 18, 2010; Revised June 11, 2010; Accepted July 08, 2010; Published August 18, 2010.

Citation: Johnson, R. K., & Petrova, E. R. (2010). A Hybrid Cryptographic Framework for Securing Cloud Storage Services Using AES and RSA. International Journal of Computer Science and Technology, 12(8), 55–61.