Summary: | 碩士 === 國立成功大學 === 電腦與通信工程研究所 === 93 === BGA (Ball Grid Array), a technology of packaging, is widely used in the integrated circuits. The model name and function of each IC chip through laser marking are to be identified by recognizing the etched mark on package surface. For instance, chip group, CPU, Flash, some communication IC etc. In order to achieve accurate alignment for the location of the etched mark, the central position and orientation angle of an IC package are examined through the image inspection system.
This thesis proposed a subpixel alignment method for BGA images. In the method, firstly, region projection is applied to recognize the contour of the object within the inspected area and delimit the regions of interest. Then the edge elements of the object can be found by using DOB (Difference of Boxes). Moreover, the edge elements resulting from cracks or flaws on object’s boundaries are filtered out by edge following method. Next, the edge elements are modified by ZOM (Zernike Orthogonal Moment) to achieve subpixel accuracy. Finally, the central position and orientation angle are obtained by the LSE line fitting algorithm followed by geometric computation.
Our proposed method is evaluated in terms of the stability and accuracy under noise degradation for synthetic images. The experimental results show that the error of the central position and the orientation is within ±0.1 pixel and within ±0.01 degree respectively without noise; the error standard deviation of the central position and the orientation are all less than 1.6% under various levels of Gaussian noise corruption. On running performance, it takes approximate 109 ms to complete the entire process for an inspected area of 360×300 pixels while running on a PC equipped with a processor of Pentium 4 3.0GHz.
|