According to the basic characteristics of LED chip image composition and gray scale inhomogeneity, a new local adaptive threshold algorithm was proposed for the segmentation of the LED grain. In the design of the threshold, characteristics of the gray level inhomogeneity of the image area and the contrast of the different points to the center point were considered.A regional contrast ratio is designed to represent this characteristic.The adaptive threshold adjustment was implemented in combination with both local mean and regional center contrast to identify the exact location of the LED grain.At the same time, in order to eliminate the unqualified LED grain and noise interference,the image morphology method based on corrosion and expansion operation was applied to filter the segmented images. The number of effective LED grain was determined by the LED chip size.The experimental results show that the algorithm has achieved good results in terms of grain counting efficiency and accuracy.