Inspection of dimensional defects has

Inspection of dimensional defects has become a critical task for manufacturers who strive to improve product quality and production efficiency [16,18]. Research concerning automated visual inspection grows rapidly. Leopold et al. [17] introduced new approaches in fast 3D-surface quality control. They made a good classification of these approaches and their application fields. Most of researches integrated AI approaches with image processing tools. Lin [12] developed an approach that utilizes neural network and statistical approach to inspect light-emitting diode chips.
Fricout et al. [7] developed an on-line approach to inspect the smoothness of metal surfaces. The measurements were based on topographical maps obtained through interferometric microscopy. The resulting data were analyzed by an algorithm based on morphological and statistical features extraction from surface topography, factorial analysis, bootstrap over-sampling and Bayesian classification. Lin [13] presented a wavelet characteristic based approach for the automated visual inspection of ripple defects in the surface barrier layer chips of ceramic capacitors.
Among the useful applications of visual inspection is the colour unevenness testing like in LCD monitors which is not easy to be made by humans as a result of human subjectivity and eye fatigue. Chiu and Lin [4] developed a hybrid approach to test blemishes in LCD panels. Their approach is based on Hotelling statistics and image analysis. Other examples includes [1,3,27,29].

The developed approach
When an image is digitally transformed and acquired by a computer, it order GSK2126458 is represented by a two-dimensional matrix. The matrix elements values depend on the type of the image that can be classified as binary, colour, gray scale, etc. As by Gonzalez et al. [8], an image may be defined as a two dimensional function, f(x,y), where x and y are spatial coordinates, and the amplitude of f at any pair of coordinates (x,y) is called the intensity or gray level of the image.
The next step is to threshold the image and transform its matrix elements into binary values (0,1). Final coordinates of masks are specified based on the clusters of zeros and ones that determine the locations of quality characteristics in the image matrix. A final step is to calculate an index for each quality characteristic that reflects the extent of conformance of that quality characteristic to the target or nominal characteristic of the reference image (the corresponding characteristic with nominal dimensions). The index is calculated for each mask (m) using the following equation:where is a (0,1) variable; which is one when the mask’s i pixel is similar to the corresponding one of the reference image and zero otherwise, is the number of pixels of mask m, and M represents the number of masks, or number of quality characteristics.
The M indices are calculated simultaneously. Each pixel of a product’s image matrix is compared to its corresponding one of the reference image; and according to its location (the mask zebroid belongs to), the index value of that mask is updated. The value of an index is exactly equal one (theoretically) if the quality characteristic dimensions are the same as the reference one and has no shift in position. Deviation from that target downsizes the index value. Experimentally, one can determine an acceptable value lower limit for when the dimensions of the quality characteristic fall within the tolerance region. The approach can also be used to detect a shift in the quality characteristic position. If, for example, a hole’s centre is shifted from its correct position, the calculated index will go down even if the hole diameter is within tolerance.
Calculating the acceptable value for different characteristics, a product can be determined whether there is a problem(s) with one or more of its quality characteristics one by one.