Accepted Papers


  • Combining Wave Atom Transform and Probabilistic Neural Network in Brain Tumor Classification
    Bilal Shehada and Mohammed Alhanjouri, Islamic University of Gaza (IUG), Palestine.
    This paper A Brain Tumor Detection and Classification System has been designed and developed using Probabilistic Neural Network combined with wave atom feature extraction. The conventional method for medical resonance brain images classification and tumors detection is by human inspection. Computerized Tomography and Magnetic Resonance images contain a noise caused by operator performance which can lead to serious inaccuracies in classification. The use of Neural Network techniques shows great potential in the field of medical diagnosis. in this paper the Probabilistic Neural Network with Wave Atom Transform was applied for Brain Tumor Classification. Decision making was performed in two steps, i) Feature extraction using the Wave Atom Transform ii) Dimensionality reduction using Gray Level Co-occurrence Matrix (GLCM). iii) Classification using Probabilistic Neural Network (PNN). The main feature extracted from shape features and texture features extracted from the disease images. Evaluation was performed on image data base of 1500 MRI Brain Tumor images collect from 14 patent.

  • Modified Hierarchical Model of Object Recognition Based on Features of Intermediate Complexity
    Saeide Hatamikhah and Jamshid Shanbehzadeh, University of Kharazmi, Iran
    The human visual system analyzes shapes and objects in a series of stages in which stimulus features of increasing complexity are extracted and analyzed. Elementary features like bars and edges are processed in earlier levels of visual pathway and as far as one goes upper in this pathway more complex features will be spotted. It has been shown that intermediate complexity (IC) features are optimal for the basic visual task of classification. Here we extended a hierarchical model, which is motivated by intermediate complexity, for different object recognition tasks. In this model, a set of object parts, named patches, extracted base on intermediate complexity. These object parts are used for training procedure in the model and have an important role in object recognition. Our reported results indicate that these patches are more informative than usual random patches. We demonstrate the strength of the proposed model on a range of object recognition tasks. The proposed model outperforms the original model in diverse object recognition tasks. It can be seen from the experiments that selected features are generally particular parts of target images. Our results suggest that selected features which are parts of target objects provide an efficient set for robust object recognition.

  • Image Processing Based Girth Monitoring and Recording System for Rubber Plantations
    Chathura Thilakarathne, Padmika Bhanusri, Tharindu Randeny, Harsha Rupasinghe and Chulantha Kulasekere, University of Moratuwa, Sri Lanka
    Measuring the girth and continuous monitoring of the increase in girth is one of the most important processes in rubber plantations since identification of girth deficiencies would enable planters to take corrective action to ensure a good yield from the plantation.

    This research paper presents an image processing based girth measurement & recording system that can replace existing manual process in an efficient & an economical manner.

    The system uses a digital image of the tree which uses the current number drawn on the tree to identify the tree number & its width. The image is threshold first & then filtered out using several filtering criterion to identify possible candidates for numbers. Identified blobs are then fed to the Tesseract OCR for number recognition. Threshold image is then filtered again with different criterion to segment out the black strip drawn on the tree which is then used to calculate the width of the tree using calibration parameters. Once the tree number is identified & width is calculated the girth the measured girth of the tree is stored in the data base under the identified tree number

    The results obtained from the system indicated significant improvement in efficiency & economy for main plantations. As future developments we are proposing a standard commercial system for girth measurement using standardized 2D Bar Codes as tree identifiers

  • Template Matching Based Non-Local Means Prediction for Hevc Intra Coding
    Xin Huang1 and Haitao Yang2, 1Technical University of Denmark, Denmark and 2Central Research Institutes, Huawei Technologies, China
    This paper proposes a Template Matching based Non-Local Means prediction scheme for High Efficiency Video Coding (HEVC). As a supplemental prediction for conventional intra prediction in HEVC, the proposed TM-NLM prediction utilizes the non-local correlation from decoded pixels and effectively improves prediction accuracy. Integrated the proposed scheme in HEVC reference software, it shows that coding gain over HEVC intra coding are 8.3% on average (up to 19.2%) for screen content sequences and 1.1% on average for natural sequences.

  • Information and Analytical Support Procurement Monitoring for the Purposes of Public Control
    Natalya Mamedova and Alexandra Baykova, Moscow State University of Economics, Russian Federation.
    Article focuses on the automation of processes of social control in public procurement. Information and analytical support procurement monitoring allowing activities to public scrutiny in the planning, allocation of procurement, contract execution.

  • Subset Selection for Landmark Modern and Historic Images
    Heider K. Ali and Anthony Whitehead, Systems & Computer Engineering Department,Canada and Carleton University, School of Information Technology, Canada.
    An automatic mechanism for the selection of image subset of modern and historic images out of a landmark large image set collected from the internet is designed in this paper. This selection depends on the extraction of dominant features using Gabor filtering. These features are selected carefully from a preliminary image set and fed into a neural network as a training set. The mechanism collects a raw large set of landmark images containing modern and historic images and non-landmark images as well, process these images, and finally classify them as landmark and non-landmark images. The classification performance highly depends on the number of candidate features of the landmark.


Copyright ® SIGL 2015