Informatics Publishing Limited

Sunil S. Harakannanavar ¹, K. S. Prabhushetty ¹, Chaitra Hugar ², Ashwini Sheravi ², Mrunali Badiger ², Prema Patil ²

Affiliations
1 Department of Electronics and Communication Engineering, S. G. Balekundri Institute of Technology, Belagavi-590010, Karnataka, IN
2 S. G. Balekundri Institute of Technology, Belagavi-590010, Karnataka, IN

Abstract

The iris pattern is an important biological feature of human body. The recognition of an individual based on iris pattern is gaining more popularity due to the uniqueness of the pattern among the people. In this paper, the iris images are read from the database and preprocessing is performed to enhance the quality of images. Further the iris and pupil boundaries are detected using circular Hough transform and normalization is performed by using Dougman’s rubber sheet model. The fusion is performed in patch level. For performing fusion, the image is converted in to 3x3 patches for mask image and converted rubber sheet model. Patch conversion is done by sliding window technique. So that local information for individual pixels can be extracted. The desired features are extracted by block based empirical mode decomposition as a low pass filter to analyze iris images. Finally the matching between the database image and test image is performed using Euclidean Distance classifier. The experimental results shows 100% accuracy on CASIA V1.0 database compared with other state-of-art methods.

Keywords

Hough Transform, Normalization, Localization, Euclidean Distance, Dougman’s Rubber Sheet Model.

Full Text

   

Shridhar H. ¹, Sumalata M. V. ², Thushara M. ², Ashwini D. N. ², M. Suma ², R. Premananda ², Sunil S. Harakannanavar ³

Affiliations
1 Department of Electronics and Communication Engineering, IN
2 Department of Electronics and Communication Engineering Government Engineering College, Haveri-581110, Karnataka, IN
3 Department of Electronics and Communication Engineering Nitte Meenakshi Institute of Technology, Yelahanka, Bangalore-560064, Karnataka, IN

Abstract

The proposed model is focused on achieving high accuracy and real-time performance in object detection using a deep learning-based approach. The two types of state-of-the-art methods for object detection were discussed: onestage methods prioritizing inference speed, such as YOLO, SSD, and RetinaNet, and two-stage methods prioritizing detection accuracy, such as Faster R-CNN, Mask R-CNN, and Cascade R-CNN. The Faster R-CNN and SSD have better accuracy, while YOLO performs better when speed is given preference over accuracy. The proposed model uses a deep learning-based approach that combines SSD and MobileNet to efficiently implement detection and tracking. The SSD eliminates the feature resampling stage and combines all calculated results as a single component, while MobileNetV2 is a lightweight network model that uses depth-wise separable convolution to perform efficient object detection without compromising on performance. The model aims to elaborate on the accuracy of the SSD object detection method and the importance of the pre-trained deep learning model MobileNetV2. The experiments were conducted on the COCO dataset to recognize objects, and the model was also tested on real-time images for object recognition. The resulting system is fast and accurate, making it suitable for applications that require object detection.

Keywords

RetinaNet, Object recognition, object detection, convolution network, MobileNetV2.

Full Text