This paper attempts to improve the commonly used image multi-focus fusion methods, for they could not identify meaningful image features from noises. An antinoise multi-focus image fusion algorithm is presented. The improved adaptive block-based image fusion algorithm combined with a new focus measure with noise immunity is used to focus the noisy image effectively and to achieve good fusion results. Root Mean Square Error (RMSE) and Mutual Information (MI) are selected to evaluate the fused noisy image with different intensities and comparison experiments are performed.As compared with those of contrast pyramid, wavelet transform and Contourlet transform, the average RMSE of the fused image by the proposed method has been decreased by 4.288 9, 4.479 1 and 4.187 1 respectively, while the average MI increased by 2.366 4, 3.282 5 and 2.063 9, respectively. With the noise interference, the proposed method can maintain the useful information of the source images accurately, suppress noise effects effectively and obtain better image fusion quality.