Date of Award

2010

Publication Type

Doctoral Thesis

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

First Advisor

Majid Ahmadi

Second Advisor

Maher A. Sid-Ahmed

Keywords

Applied sciences, Artificial neural network, Biomedical imaging, Document recognition, Process control

Rights

info:eu-repo/semantics/openAccess

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Abstract

In computer-vision based system a digital image obtained by a digital camera would usually have 24-bit color image. The analysis of an image with that many levels might require complicated image processing techniques and higher computational costs. But in real-time application, where a part has to be inspected within a few milliseconds, either we have to reduce the image to a more manageable number of gray levels, usually two levels (binary image), and at the same time retain all necessary features of the original image or develop a complicated technique. A binary image can be obtained by thresholding the original image into two levels. Therefore, thresholding of a given image into binary image is a necessary step for most image analysis and recognition techniques. In this thesis, we have studied the effectiveness of using artificial neural network (ANN) in pharmaceutical, document recognition and biomedical imaging applications for image thresholding and classification purposes. Finally, we have developed edge-based, ANN-based and region-growing based image thresholding techniques to extract low contrast objects of interest and classify them into respective classes in those applications.

Real-time quality inspection of gelatin capsules in pharmaceutical applications is an important issue from the point of view of industry's productivity and competitiveness. Computer vision-based automatic quality inspection and controller system is one of the solutions to this problem. Machine vision systems provide quality control and real-time feedback for industrial processes, overcoming physical limitations and subjective judgment of humans. In this thesis, we have developed an image processing system using edge-based image thresholding techniques for quality inspection that satisfy the industrial requirements in pharmaceutical applications to pass the accepted and rejected capsules.

In document recognition application, success of OCR mostly depends on the quality of the thresholded image. Non-uniform illumination, low contrast and complex background make it challenging in this application. In this thesis, optimal parameters for ANN-based local thresholding approach for gray scale composite document image with non-uniform background is proposed. An exhaustive search was conducted to select the optimal features and found that pixel value, mean and entropy are the most significant features at window size 3x3 in this application. For other applications, it might be different, but the procedure to find the optimal parameters is same. The average recognition rate 99.25% shows that the proposed 3 features at window size 3x3 are optimal in terms of recognition rate and PSNR compare to the ANN-based thresholding technique with different parameters presented in the literature.

In biomedical imaging application, breast cancer continues to be a public health problem. In this thesis we presented a computer aided diagnosis (CAD) system for mass detection and classification in digitized mammograms, which performs mass detection on regions of interest (ROI) followed by the benign-malignant classification on detected masses. Three layers ANN with seven features is proposed for classifying the marked regions into benign and malignant and 90.91% sensitivity and 83.87% specificity is achieved that is very much promising compare to the radiologist's sensitivity 75%.

Share

COinS