Rectangular And Circular Objects Recognition In PCB-AXI Images

Automatic X-ray Inspection(AXI) plays an extremely important role in quality control which is the most important factor in PCB automatic manufacturing.Rectangular and circular objects recognition is the key in AXI. The proposed rectangular and circular object recognition algorithms can be obtained by judging the geometric characteristics that is extracted based on the analysis of Hough Transform which has a high computing cost, making it unable to satisfy the instantaneity requirement of the system. During the detection of circular objects, the proposed methods can lead to the interruption of complicated images, inevitable mistakes and omissions. Thus, unable to acquire the contours of the circular objects. In AXI applications, shape angle theory has been proven as an effective method to detect rectangular area in data consisting of discrete points by calculating the tangent of discrete points. The current methods based on Vialard algorithm will increase the complexity and lead to some non-neglectable errors because of multiple iterational calculations. Due to the existence of large-area shelters in complex images, the segmentation results obtained by global threshold are not ideal and the intensity gradient will be complicated. The recognition results are not ideal while applying original HT methods and the normal vector of the PCB will not parallel to the direction of the beam. More details can be found in the image under micron level of resolution.For example, interruption on the solid ball is formed by wires and other objects, which cannot be removed by using proposed methods. In addition, the time complexity of the recognition algorithm has to be controllable in the AXI system applications.In this paper, a new method based on Fourier fitting has been proposed to solve the rectangular objects recognition problem. Wherein, a polar coordinate system transform has been applied to the discrete points at first. Then Fourier series are used to fit envelop and calculate the derivative of the envelop function to get the tangent. This method avoid the iterations in many methods based on Vialard algorithm, so time complexity degrade from O(n3) to O(n). Finally, an algorithm for recognition of rectangular object based on proposed method is introduced. Experiments have been implemented as well to verify the correctness and reliability. For example, the method applied in this paper consumes 1.5775 s in average, while it is 156.155 s when usingVialard method. The method applied in the paper also produces more accurate results.To solve the problem of BGA solid ball contours extraction in complex images, a solid ball extraction method based on subimages is applied: To acquire the position of all the solid balls including the undetected ones in BGA images, the distribution of the grid and original HT method is implemented, therefore the image can be divided into subimages where every subimage contains a single solid ball as ROI for further processing;Mathematics modeling is implemented on the X-radiation of BGA imaging process to determine the intensity distribution. Every solid ball is processed in accordance with this evidence, thus the binary image that is obtained contains majority of the wanted contours. Finally, an adaptive interruption exclusion algorithm based on sliding variance of the histogram is obtained from the values of the polar coordinate system. By utilizing the geometric characteristics of the grid-ranged solid balls, the interfered points in the subimage can be get rid of and be replaced with arcs. The contour extraction is then completed.The rectangular objects recognition algorithm proposed in this paper can be applied to avoid the problem of over iteration created by using Vialard algorithm and its derived methods. The method proposed in this paper reduces time complexity by two orders and produces more accurate results. The improved shape angle calculation algorithm is applied to rectangular contours recognition. By case analysis, the accuracy and reliability of this method is being demonstrated. The proposed BGA solid ball contours extraction algorithm is one extension of the current algorithm, making it possible to solve the problem that could not be solved by using the current algorithm. It also provides the theoretical evidence for the threshold value selection during the conversion of binary BGA image.