Image fusion for a digital camera application based on wavelet domain hidden Markov models

The traditional image fusion for a digital camera application may not be satisfactory to classify pixels in the source image by the statistical techniques. In this paper, we present a technique, based on wavelet domain hidden Markov models (HMMs) and max-likelihood estimation. The method presented here consists of deciding the quality of pixels in source images directly from the statistical techniques to overcome the shift-variant of inverse wavelet transform. We have studied several possibilities including all energy methods that are utilized in the standard image fusion for digital camera application and discuss the difference to our new method. The new framework uses two trained HMMs to decide if the wavelet coefficients of source images are in-focus or out-focus, and then judges the quality of pixels in the source images directly to overcome the shift-variant of inverse wavelet transform. The two trained HMMs are obtained separately from an in-focus image and a out-focus image using EM algorithm. We used the method to merge two images in which have two clocks with different focus and obtain the best fusion results in preserving edge information and avoiding shift-variant.