Abstract: To solve the problem of feature point detection and matching in the image registration techniques at present, an improved image registration algorithm was proposed based on block information entropy and characteristic scale. By dividing the image into blocks and combining the information entropy of each block, the concentrated distribution problem of feature points extracted by the Harris-Laplace operator was improved. By comparing the value of corner response function, redundant points in the feature points were eliminated. By combining the scale information of feature points with Hu moments and bidirectional matching strategy, the accuracy of the initial matching points was enhanced. Simulation shows that the improved registration algorithm can realize high-precision image registration and possesses strong robustness with the geometric transformation of image.