Digital Image Processing For Defect Recognition: Design Specification And Approach

Discuss how the digital image processing uses computer algorithms to carry out image processing tasks on digital image files?

Digital image processing uses computer algorithms to carry out image processing tasks on digital image files. This is more related to the domain of digital signal processing. Compared to analog image processing, there are various advantages of digital image processing and it has wide range of application in various domains. Digital image processing can be used to find out noise and distortions in digital images. An image can have two or more dimensions. Digital processing can be also extended to multi-dimensional models. (Burger & Burge, 2009), (Jähne, 2005).

Large building, civil infrastructures like pipelines are prone to damages over time. Digital image processing can be used to find such damages in a faster and effective way. It will help state agencies to monitor their assets and take decision about managing those assets. Thus this is an application of digital image processing in civil engineering domain. The applications of digital image processing is more focused on areas like inspection on underground pipelines, conditions of pavements on roads etc. (Zheng, Kong, & Nahavandi, 2002)

In the next sections of this report, there will be discussion on design specification and approach taken in designing a model for detecting defects on metal surfaces and rusts.

Any automated defect recognition method based on digital image processing will have some common stages. Those stages are,

1. Acquisition of image
   
   
2. Processing of image
   
   
3. Analysis of data.

In this stage, images of the objects are acquired. Automated defect recognition system based on digital processing are mainly used by bridges and other civil constructions, on such constructions a common damage is rust, breaks etc. Digital image processing helps in detecting these damages. Digital images of the damages are acquired. These pictures are taken manually and retaining quality and visibility of the parts, color, quoting etc. of the bridges. After taking the images, data sets are created from those images. These data sets will be used in later part of the process and in data analysis. These sets are further tested and broken down into two groups consisting of defective and non-defectives images. Images of the bridge parts having no rust are kept in the non-defective images group. Images with small to medium levels of rusts are kept into defective group. (Lee, 2010)

In this stage, the color images are transformed into greyscale images. There are three primary colors in a color image. The primary colors are red, green and blue. Any color can be developed from different combinations and quantity of these three primary colors. There are total 224 possible colors that can be developed by mixing these colors and under 256(28) color shades. Images represented by grey scale will be represented by 8 bits only (Xie, 2007). Based on the lights intensity a value will be assigned to each of the light intensity representation ranging from 0 to 255. Each pixel will have any of these 256 values. 0 represents black and 1 represents white. All other values represents some variations of grey between black and white. It will help to reduce the size of images significantly. Thus the computational efficiency while working with the images will be improved. Now, a defect image will be compared to another defect and non-defect images. The level of similarity and dissimilarity between the results from these two comparison will be calculated. (Lee, 2010)

Acquisition of Image

A method to compute these comparisons is representing the images using matrices and then comparing the Eigen values of the matrices. If there is a larger difference in the pair wise comparison then the differentiating power will become significantly higher.

The calculation process goes as, two dimensional special coordinates are used for representing digital images. For example, f(x,y) will be of size m x n. at any point (x,y) the value of f(x,y) will be proportional to the brightness of that point in the image. In this process, pixels with more brightness are assigned with higher values and pixels with more darkness are assigned to some lower value. Then one reference image of same size is added to the process. Then the covariance matrix is calculated. The eigenvalues of this matrix is used for extracting shape information from the grey scale distribution from the pair wise comparisons. If the eigenvalue is larger enough then that represents the variance along the major axis of the shape of the pixel distribution. If the eigenvalue is smaller enough then that represents the variance along with the minor axis of the same. A two dimensional mapping is used for representing the distribution. If there is two identical images compared in the pair wise comparison then the smaller eigenvalue will be near to or equal to zero.  (Louban, 2009)

At this stage, images of the coating of the objects are processed for generating Eigen values. Again two pairwise comparisons are done. One is about comparing two non-defective images and another is -comparison of one defective image with another non-defective image. All results from these comparisons are obtained. Five attributes are calculated from the results. These five attributes are average, maximum, minimum, variance and standard deviation. Then a scale is developed for categorizing and calculating the Eigen values.

There are various approaches in automated defect recognition method by using digital image processing. Some of the earlier works were based on mathematical calculations based on matrices, statistical analysis etc. With time digital image processing technologies have advanced and these defect detection algorithms, process, procedures have also advanced. In this work, basic mathematical approach based on Eigen values of the matrices, co-variance etc. have been used. (Lee, 2010), (Mery & Rueda, Advances in Image and Video Technology, 2007)

There are basically three important parts of the system. First part is a good digital image processing unit that will be able to taking good quality images of the objects. In most of the cases automated defect recognition based on digital image processing procedures are used for bigger civil constructs like bridges, pipelines etc. Thus taking good quality pictures from right angels and distances is necessary.

After designing the image processing system, it needs an application that will convert the color images into appropriate greyscale formats. It should be done very carefully by focusing on the range of greyscale representations of the pixels. (Lee, 2010)

The application also needs to apply transforming the greyscale images into corresponding matrix forms suitable for further analysis and then calculation of the matrices, Eigen values, co-variances etc. From these calculations, the representations will be compared with two different images from the defect and non-defect groups. The comparison module will find the differences between the pairs.

Processing of Image

Based on the values found from the comparisons statistical analysis processes will be applied to find more useful information about the defects, levels of defects and possible suggestions like minimum defect, maximum defect, average defect etc. Based on these suggestions, further actions on those damages can be decided.

Result and Conclusion

In this discussion so far, the focus has been given on automated defect recognition based on digital image processing for the civil constructs like bridges, pipelines etc. in most of the cases damages to these constructs are related to rust, fracture etc. For finding suitable methodology and design approach, literature review of the previous works have been used. There have been significant amount of work on this topic. There are various methodologies and approaches are available from those the one based on matrices, Eigen value and covariance has been selected. This process is particularly helpful for recognizing any rust on the coating of the bridges through image processing activities. The defect recognition method or process has been designed based on pair wise comparison of images and calculation of Eigen values. In the selected approach Eigen values is a key distinguishing feature between a defective image and a non-detective one. (Lee, 2010), (Mery, 2002)

The methodology used for detection of rust, is based on three stages, acquisition of images, processing of images and analysis of data found from processing of images. During acquisition of images, images of the object like a bridge, are taken and those are separated into two data sets. The data sets are defective images and non-defective images. In the next stage, processing of those images are carried out. Processing is mainly related to transforming the color images into grey scale images then calculating Eigen values of the image and then comparing the Eigen values for images from each of the pairs. The Eigen values found from the analysis are distributed on some two dimensional distribution map. Five statistical values are then calculated from the distribution and those are categorized into a suitable tabular format. This method is effective for identifying rusts on bridges etc. as described in the paper. But there are some limitations of these methods and approaches. External characteristics of an object can be easily and effectively captured by digital image processing methodology. But examination of the internal conditions is not possible using digital image processing. This is more complex when an object is like pipelines etc. internal damages to the pipelines is not possible in this process. So, this method is only acceptable if external condition examination is primary importance. In practice, comprehensive field testing process is needed along with these methodologies. It will enhance validity of the proposed methodology. (Lee, 2010)

References

Burger, W., & Burge, M. J. (2009). Digital Image Processing. Springer .

Castleman. (2007). Digital Image Processing. Pearson .

Jähne, B. (2005). Digital Image Processing. Springer.

Kamel, M., & Campilho, A. (2007). Image Analysis and Recognition. Springer .

Lee, S. (2010). Automated Defect Recognition Method by Using Digital Image Processing. Conference on Associated Schools of Construction. Boston.

Louban, R. (2009). Image Processing of Edge and Surface Defects. Springer .

MacKenzie, D. S., & Totten, G. E. (2005). Analytical Characterization of Aluminum, Steel, and Superalloys. CRC Press.

Mery, D. (2002). New Approaches for Defect Recognition with X-ray Testing.

Mery, D., & Rueda, L. (2007). Advances in Image and Video Technology. Springer .

Toriwaki, J., & Yoshida, H. (2009). Fundamentals of Three-dimensional Digital Image Processing. Springer .

Xie, X. (2007). A Review of Recent Advances in Surface Defect Detection using. Electronic Letters on Computer Vision and Image Analysis, 1-22.

Zheng, H., Kong, L., & Nahavandi, S. (2002). Automatic inspection of metallic surface defects using genetic algorithms. Elsevier.