Chandana Dinesh Parape
Kyoto University
Network
Latest external collaboration on country level. Dive into details by clicking on the dots.
Publication
Featured researches published by Chandana Dinesh Parape.
International Journal of Machine Learning and Cybernetics | 2015
Chinthaka Premachandra; H. Waruna H. Premachandra; Chandana Dinesh Parape; Hiroharu Kawanaka
We present an image analysis based automatic road crack detection method for conducting smooth driving on non-smooth road surfaces. In the new proposal, first the road surface areas which include cracks are extracted as crack images by analyzing the color variance on norm. Then cracks are extracted from those areas by introducing a method on discriminant analysis. According to experiments using the images of different road surfaces, the new proposal showed better performances than the conventional approaches.
international geoscience and remote sensing symposium | 2011
Chandana Dinesh Parape; Masayuki Tamura
This paper presents a methodology and results of evaluating damaged buildings detection algorithms using an object recognition task based on Differential Morphological Profile (DMP) for Very High Resolution (VHR) remotely sensed images. The proposed approach involves several advanced morphological operators among which an adaptive hit-or-miss transform with varying size, shape and gray level of the structuring elements. IKONOS satellite panchromatic images consisting of a pre and post-earthquake site of the Sichuan area in China were used. Morphological operations of opening and closing with constructions are applied for segmented images. The Unsupervised classification ISODATA algorithm is used for the feature extraction and the results comparison with ground truth data, complex urban area before the earthquake gives 76% and same area wracked after the earthquake gives 88% buildings detection on object based accuracy.
international geoscience and remote sensing symposium | 2012
Chandana Dinesh Parape; H. Chinthaka N. Premachandra; Masayuki Tamura; Masami Sugiura
The 2011 Great East Japan Earthquake occurred 130km off the northeast coast of Japan in the Pacific Ocean near the Japan Trench and it triggered huge Tsunami, which caused damaged along 600km of coastline. Estimation of tsunami impact on buildings from this event is important for quick response for evaluation and the reconstruction process. This study presents a methodology for an automated damaged buildings detection algorithm using an object recognition task based on Differential Morphological Profile for Very High Resolution (VHR) remotely sensed satellite images. The proposed approach involves morphological operators among which adaptive varying specific sizes, shape and gray level of the structuring elements. GeoEye-1 and IKONOS satellite images consisting of a pre and post 2011 Pacific coast Tohoku earthquake and tsunami site of the Ishinomaki area in Miyagi Prefecture, Japan were used. Morphological operations of opening and closing with construction were applied for segmented images. The Random Forest classification method was used for the building extraction and the results comparison with ground truth data.
International Journal of Machine Learning and Cybernetics | 2017
H. Waruna H. Premachandra; Chinthaka Premachandra; Chandana Dinesh Parape; Hiroharu Kawanaka
This paper presents a speed-up ellipse enclosing character detection algorithm that uses parallel image scanning and the Hough transform (HT) for large-size document images. Objects in images are generally detected based on geometrical information obtained via raster scanning. In raster scanning, all pixels of an image are scanned starting from the upper-left point and ending with the lower-right point. In the case of large-size images, considerable time is needed for processing an image by scanning all pixels. In this paper, an object detection approach for large-size images is proposed which does not require scanning all pixels in the images. In this speed-up ellipse enclosing character detection approach for large-size document images, pixels are scanned on constantly spaced vertical parallel lines. If an object larger than a certain size is identified while scanning, the presence of an ellipse enclosing character is assumed and ellipse detection is conducted by applying HT only in a defined local image area over the found object. With this approach, processing time can be dramatically reduced by disregarding some objects and reducing the total image area used for ellipse detection.
Special Session on Computer VISION for Natural Human Computer Interaction | 2016
H. Waruna H. Premachandra; Chinthaka Premachandra; Chandana Dinesh Parape; Hiroharu Kawanaka
Hough transform (HT) is typically used to detect lines in images, but that method is slow due to its use of voting-based parameter detection; detecting lines in large document images can take dozens of minutes. Nonetheless HT is very effective at detecting lines, so we investigate methods for fast HT-based line detection of large document images by minimizing Hough space processing and reducing the image area used for line detection with parallel pixel scanning and local image domain analysis. We conduct experiments to confirm the effectiveness of the proposed method using appropriate large documents images. The results show a significant computational time reduction as compared to conventional methods.
international conference on image analysis and recognition | 2014
Chinthaka Premachandra; H. Waruna H. Premachandra; Chandana Dinesh Parape; Hiroharu Kawanaka
A fast dot/dash line detection method suitable for large scale binary document images is proposed. The method works by reducing the number of scanned pixels used for the detection process. In the new method, pixels in vertical image layers with only a constant spacing are scanned. By using this technique, the computational time can be reduced because some of the uninteresting objects in the image can easily be omitted in the scanning stage. The new method is faster than the conventional method not only due to its scanning method but it also due to the simple process used for detecting dot/dash lines. A dot/dash line is detected by selecting a small defined image domain from the large scale image. We evaluated the new method against conventional methods on appropriate document images and found an improved processing time without any significant loss of line detection ability.
systems, man and cybernetics | 2013
Chinthaka Premachandra; H. Waruna; H. Premachandra; Chandana Dinesh Parape
We present an image based automatic road crack detection method for achieving smooth driving on deformed roads. First, the road areas which include cracks are extracted as crack images analyzing the pixel variance of the road image. Then cracks are extracted from crack images by introducing a method on discriminant analysis. According to experiments using different road images, the new proposal is effective in detecting road cracks.
International Journal of Machine Learning and Cybernetics | 2015
Chandana Dinesh Parape; Chinthaka Premachandra; Masayuki Tamura
情報科学技術フォーラム講演論文集 | 2013
Chinthaka Premachandra; H. Waruna H. Premachandra; Chandana Dinesh Parape
Sustainability | 2013
Chandana Dinesh Parape; Chinthaka Premachandra; Masayuki Tamura; Abdul U. Bari; Duminda Welikanna; Shengye Jin; Masami Sugiura