A New Eight-Connected Components Analysis Algorithm for Super-large Binary Image Using CUDA

Starting from the storage and transmission of features, this paper proposes an improvement plan to solve the time-consuming problem of data transmission of the connected components analysis algorithm in industrial surface defect detection applications. This solution is based on the characteristics of eight-connected components, and greatly improves the efficiency of connected components analysis through compressed storage space and sparse matrix transmission processing. In the experiment, we tested three different sizes of binary images, and the results showed that as the image size increases, the algorithm performs better. This is because the space occupied by the effective connected components features is much smaller than the space for the maximum number of connected components applied in advance. In the uploading and processing stages, the time consumption is basically the same as that of the original algorithm, but in the downloading stage, the time cost has been significantly improved. This improvement will greatly improve the efficiency of practical industrial defect detection applications.