Image compression is an indispensable process in data transmission and storage. The fractal image compression method has unique advantages due to its simple compression method, reconstruction at any scale, fast decoding speed and high compression ratio. However, the traditional fractal image compression method has a defect that the encoding time is too long. Aiming at the imbalance between the compression ratio and the recovery effect, it is necessary to solve the problem of too long time in the encoding process under the premise of ensuring the image restoration effect. So a fractal image compression method based on centroid feature and important sensitive area classification is proposed. By constructing the centroid feature, the problem of searching the minimum Mean Square Error (MSE) of the   block in the basic fractal is converted into the problem of searching the best matching block of the   block centroid feature in the corresponding   block centroid feature codebook. It simplifies the block search process, changes a global search to a local search, considers the important sensitive area of the image, and adopts a global search for important sensitive areas, thereby increasing the visual effect of the restored image. Experimental simulation shows that, compared with the basic fractal image compression algorithm, the centroid feature method can effectively shorten the coding time. Under the premise of achieving a satisfactory image restoration effect, this method can save the coding time by about 64% compared with the basic algorithm. This method can achieve better recovery effect than the sum of double cross/eigenvalues methods.