As the focusing range of optical imaging system is generally limited, it is difficult to make all the objects of the same scene clearly shown in one image. Besides, a case usually rose that the fused image with a high entropy, however, is not satisfying for vision effect. In this paper, a new method of multi-focus image fusion based on adaptive dividing blocks using comprehensive index was proposed, in which the comprehensive index was on basis of spatial frequency and entropy. The comprehensive index is better with the higher spatial frequency and entropy. Firstly, the registered original images were divided into a series of blocks of which the sizes were proper and the same, and then the comprehensive index for each block of source images was calculated as the focus criterion function to select an optimal block for each corresponding block of the fused image. In view of the relevance between pixel and pixel in one image, the optimal blocks selected were fused with a global fusion function. Furthermore, the sum-modified-Laplacian of fused image was used as the measure function to supervise the adaptive blocking, in which the optimal block was obtained when SML of the fused image had reached a high value or the iteration had achieved the specified numbers. Finally, the optimal size of the sub-block was automatically obtained, which was used to fuse the source images. As it was shown in the experimental results, the proposed method which was simple, but more effective compared with the traditional multiscale decomposing methods such as wavelet transform, wavelet packet transform, contourlet transform and so on. At the same time, the proposed method was also superior to the method in the literature for it could remove boundary discontinuities between image blocks. Contemporarily, much more details and edges information of the source images were reserved in the fused image.
This paper presents a method for multi-exposure images fusion based on wavelet packet transform, combining the local energy distributions of multi-exposure images with the edge detection. After decomposing two images involved in fusion into sub images in low-frequency and high-frequency with wavelet packet transform, we use different methods for low-frequency and high-frequency to obtain fusion coefficients. In low frequency processing, the method that threshold value is set for local energy is used while the edge detection method is used in high frequency, where the edge detection operator help compute the information quantity of different high frequency images. Then the coefficients for fusion are selected according to different strategies adopted for low- and high-frequency. Finally, the fusion image is reconstructed through inverse wavelet packet transform. The result shows that the fusion method is effective and the fusion image can preserve the details of the each input image successfully.