Digital image recognition methods for infrastructure surface coating assessment

Po-Han Chen, Purdue University

Abstract

As computerized technologies were widely utilized, digital image processing was also prevalently adopted in industries. In the construction area, image processing has been used for defect recognition on steel bridge painting and underground sewer systems. However, non-uniformly illuminated images always cause recognition problems and affect the accuracy. In order to resolve these problems, the neuro-fuzzy recognition approach (NFRA) was proposed. NFRA segments an image into three areas based on illumination and conducts area-based thresholding. A neural network is used in this approach for automatic generation of three optimal gray level thresholds, with the three average illumination values of the three segmented areas as the input. A fuzzy adjustment system is utilized to smooth and adjust the gray levels of the image pixels along the boundaries between different areas. After the three optimal thresholds are available and the fuzzy adjustment is done, the image can be thresholded based on the three segmented areas. The evaluation of NFRA will be made by comparing NFRA with three other methods. The three methods are: the multi-resolution pattern classification (MPC) method proposed by Chang in 2000, the iterated conditional modes (ICM) algorithm proposed by Besag in 1986, and the neural network hybrid model (NNHM) proposed by AbdelRazig in 1999. NFRA and NNHM utilize artificial intelligence (AI), such as neural networks and fuzzy systems; MPC and ICM are statistical approaches. All these methods process images in gray scale. The four methods will be compared using both synthetic spot images and rust images. Finally, conclusions will be drawn and recommendations for future work will be made.

Degree

Ph.D.

Advisors

Chang, Purdue University.

Subject Area

Civil engineering

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS