Aiming at the problem of slow algorithmic stitching and low accuracy due to the existence of a large number of similar and neatly arranged LEDs in the existing microscopic images of Micro-LED chips， this paper proposed a clustering-based Micro-LED chip micro-image stitching method. Firstly， the OTSU method was used to preprocess the images， and then the adaptive area expansion phase correlation method was used to peak the overlapping regions of two adjacent images to obtain the offset of the preliminary alignment. Subsequently， the optimized alignment strategy using image chunking and the DHash algorithm was employed to obtain the optimized offset. Finally， density-based spatial clustering of applications with noise was introduced to automatically screen out and deal with the possible erroneous offsets. The experimental results show that the average time consumption of this paper's algorithm is 64 ms， which meets the demand of real-time stitching， the stitching correct rate reaches 99.4%， the alignment accuracy is controlled within 1 pixel， and the fusion processing can achieve a perfect stitching effect subjectively， which also provides a new solution to the overall misalignment problem caused by the wrong offsets. The algorithm in this paper effectively solves the key link of microscopic image stitching in wafer-level Micro-LED chip inspection and can be promoted and applied to other machine vision automation fields with repetitive features and high precision requirements， with high practical engineering application value.