Die bonding machine is one of the primary equipments for chip production process,and machine vision inspection system plays an important role in improving the quality of die bonding machine.In the system,accurate image segmentation is a premise for image analysis.Because of spherical aberration,three algorithms of image segmentation for IC chip degraded image were analyzed,which are based on iterative algorithms,improved adaptive threshold and homomorphic filtering,respectively.Then experimental comparison results show that the improved algorithm can meet the high-precision and high-speed requirements of IC chip detection.

Because of light changing from environment in semiconductor packaging,the images from video equipment exists significant light differences.In order to reduce the affect from light differences,a multi-template algorithm based on homomorphic filtering and air space was proposed.First a rough projection method was used to determine the location of the chip,then homomorphic filtering was applied on intercepting images with space transform,thus it weakened the impact of the light change on images.Finally a template from template library was used for traditional template matching and updating the template library with interception image.Experimental results show that the algorithm of illumination has a strong self-adaptability and good application prospects.

Wavelet-structure moments for image recognition were presented to deal with the images which were translated,scaled and rotated.This new theory was based on structure moments and wavelet transformation,after changing the structure of a image function,namely the density in geometric moments,the difference between images was enlarged,using mother wavelet of the cubic B-spine,both global feature and local feature can be extracted efficiently.Experiments demonstrate that wavelet-structure moments perform better than improved Hu moments and structure moments at recognition rate and correctness.And the method was applied into practical automatic golden ball wire bonder systems.

Angle recognition for chip component is one of the most difficult issues in electronics manufacturing,and it has direct impact on the mounting precision.SUSAN based on the gray comparison but not gradient.The selection of template and threshold was discussed,after analyzing the detecting task of chip in-depth,a new algorithm based on improved SUSAN edge detecting was proposed to detect edge of chip component.Then the advantages and disadvantages of improved SUSAN and other traditional algorithm for edge detection were compared by experiments.The results demonstrate that the proposed algorithm can detect the edge of component efficiently,even under the complex illumination condition,and the precision and stability of angle recognition are both improved.