The work aims to propose a secondary matching algorithm of tablet features based on vision technology, to solve the problem of inability to efficiently and accurately count mass produced tablets due to the rapid development of the current pharmaceutical industry by means of machine vision technology. Firstly, the tablets were preprocessed and divided into multiple connected regions. Then, two tablets with large differences in shape features were selected as regions of interest with the area characteristics, and they were inflated to become the targeted prior model. The edge contour was extracted by the Canny operator, and the direction vectors of the contour points were calculated. The three-tier image pyramid search algorithm was used to speed up the matching accuracy, and the least square method was applied to adjust the matching accuracy of the template, so that the matching accuracy reached the sub-pixel level. Based on the experimental analysis of different elliptical tablets, the minimum matching score thresholds that matched templates 1 and 2 (scaling was 1:1) were respectively set as 0.63 and 0.59. With the three-tier image pyramid search algorithm, it only took 0.11 s to match the count from the creation of template. Compared with the thee-tier pyramid, the scaling was 0.7-1.0, the single-template matching algorithm speed with the minimum matching score of 0.6 was 0.07 s faster, and it could still effectively count part of the overlapped tablets with the matching accuracy of 100%. The secondary matching technology of the tablets can amplify the detection speed. The image pyramid search algorithm can be used to greatly shorten the matching time and the least square method can be applied to improve the matching accuracy of the template and increase the matching accuracy of the tablets.