Accepted Papers

  • Runway Detection using K-means clustering method using UAVSAR Data
    Ramakalavathi Marapareddy and Sowmya Wilson Saripalli ,University of Southern Mississippi Hattiesburg, MS 39406-0001, USA

    Remote sensing data gives the essential and critical information for detecting or identifying an object, a place, image fusion, change detection, and land cover classification of selected area of interest. The runway detection is an important topic because of its applications in military and civil aviation fields. This paper presents an approach for runway detection using Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVAR) data by implementing K-means clustering method. The obtained results reveal that we can obtain better detection, for the 9 and 11 classes, with iterations set to 10. In this work, the effectiveness of algorithm was demonstrated using quad polarimetric L-band Polarimetric Synthetic Aperture Radar(polSAR) imagery from NASA Jet Propulsion Laboratory°«s (JPL°«s) Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR). The study area is Louis Armstrong New Orleans International Airport, LA, USA.

  • Continuous arabic speech recognition using hmm-DNN system over mobile networks DSR and NSR based
    Lallouani Bouchakour and Mohamed Debyeche, Faculty of Electronics and Computer Science LCPTS,Algerie

    In this paper, we have investigated different systems of classification in the field of speech recognition, in order to improve its performances in client-server communication over mobile networks. To get a high correct classification rate, and compare the influence of the low bits rate coders and the efficient of dimension reduction feature vectors (Back-end) using the techniques LDA (Linear Discriminant Analysis). We have used two standards coders of ETSI (European Telecommunication Standardization Institute), the first coder is AMR-NB (Adaptive Multi Rate narrow-band), and the second coder is DSR (Distributed Speech Recognition). Our main goal is to use the different systems of classification, the CHMM (Continues Hidden Markov Models), DNN (Deep Neural Network) and HMM-DNN hybrid. Our results proved that HMM-DNN can achieve the rate accuracy almost 98.93% of corpus clean speech, 96.98% for corpus transcoded AMR-NB (5.9 Kbits/s) and 97.05% for corpus transcoded DSR (4.8 Kbits/s) compared to the model HMM 86.47% of corpus clean speech, 73.19% for corpus transcoded AMR-NB and 84.68%for corpus transcoded DSR. We can conclude that HMM-DNN gives the best results.

  • A recursive approach to detect and extract region eliminating texts from scanned land map image
    Md. Omar Faruk Rokon, Md. Rayhanul Masud,Dr. Md. Monirul Islam,Bangladesh University of Engineering and Technology, Dhaka, Bangladesh

    Elimination of unnecessary texts from land map images is the most important task in order to digitize land map from scanned map images. However, texts in land map images are a complex feature to extract, as texts lying in images are in different size, orientation, and fonts in the image. Because of these variations, detection and elimination of texts is a challenging task. Complex background and low resolution are also responsible for this difficulty. However, our approach recursively detects and extracts connected region boundary, ignoring the texts in the image which in turns produces texts free image to digitize the land information. The test result of the proposed approach on a set of scanned land map images is satisfactory.


    A new approach of brain tumor segmentation using fast convergence level set
    Virupakshappa,Department of CSE, Appa IET,India , Dr. Basavaraj Amarapur,Department of E & E E, P D A C E, Kalaburagi, India.

    Segmentation of region of interest has a critical task in image processing applications. The accuracy of Segmentation is based on processing methodology and limiting value used. In this paper, an enhanced approach of region segmentation using level set (LS) method is proposed, which is achieved by using cross over point in the valley point as a new dynamic stopping criterion in the level set segmentation. The proposed method has been tested with developed database of MR Images. From the test results, it is found that proposed method improves the convergence performance such as complexity in terms of number of iterations, delay and resource overhead as compared to conventional level set based segmentation approach.

  • A Digital Approach for Automatic Detection of Cervical Cancer through Image Analysis: A Review
  • kumar Dron Shrivastav,Ankan Mukherjee Das,Dr. Rajiv Janardhanan,Amity Institute of Public Health,Dr. Harpreet Singh,Indian Council of Medical Research,Dr. Priya Ranjan,Amity School of Engineering and Technology,Noida, Uttar Pradesh

    Cervical Cancer diagnosis is a very sensitive and crucial task. It relies heavily on the pathologist's qualitative analysis and subjective knowledge for accurate and precise diagnosis of cancer which is time exhausting. The prevalence of misdiagnosis and missed diagnosis is very high in absence of an appropriate diagnostic tool. We require an accurate and specialized expertise to confirm cancer report and that specialized expertise is not available everywhere, especially in remote areas where the patients presents with advanced clinical stage. The average waiting time to get the result for the diagnosis is one week, if the patient reaches at the right place. Digital Quantification of the cytopathological and histopathological images can reduce the burden and time of pathologist by providing accurate and clear view of microscopic images. If one handy point of care device is available even at remote places, it will increase early and quick diagnosis of cervical cancer.