Multi-focus Image Fusion Based On Multi-scale Neighborhood Technology

Due to the limited depth of optical lenses, it is hard to obtain an image that contains all relevant objects in focus. However, people want to get an image in which all objects are focused in some circumstance. In this case, more and more researchers concentrated their attentions on multifocus image fusion. With the high-speed development of information technology, image fusion have been widely applied in many field such as microbial imaging, computer vision, robot technology, digital photography, automatic target recognition, military use et al.Multifocus image fusion algorithms can be categorized into transform domain based methods and spatial domain based methods. The former techniques mainly aim at designing fusion rules for multiscale coefficients. However, the fusion rules usually suffer from common drawbacks such as:contrast reduction and image degradation, lack of considering the neighbor connection of pixel, insufficient extract focused pixels form source images, introducing irrelevant information into fused image. For the spatial domain based methods, focus region detecting methods could produce pleasing results than those of pixel based, block based, and segmentation based methods. Unfortunately, some problems such as choosing filter parameters and fusing focus border et al. impede the fusion performance of these focus region detecting methods.To overcome the defects and deficiencies exist in the traditional multifocus image fusion methods, by employing the multiscale neighbor technique, this paper proposes two efficient algorithms in transition domain and spatial domain, respectively.Aiming at responding the defects in transform domain, an efficient multifocus image fusion scheme in nonsubsampled contourlet transform domain is proposed, the contributions are two new fusion rules as follows:(1) Based on the property of optical imaging and the theory of defocused image, paper investigates the connection between a low-frequency image and the defocused image, and in view of the connection among neighbor pixels, and then presents a selection principle for lowpass frequency coefficients, (2) Considering the property of bandpass subband coefficients, proposing multiscale curvature based method, which not only inherits the advantages of windows with different sizes. but also correctly recognizes the focused pixels from source images, and then develop a new fusion rule to fuse the bandpass subband coefficients.To address the weakness of focus region detection algorithms in spatial domain, paper proposes a new method that combining mixed-order structure tensors and multiscale technique. The contributions contains:(1) Development of a new focus measure that combines the fractional structure tensor and the integer order structure tensor, which can effectively avoid erroneous assessments of the focusing properties when the focused pixels are located in smooth regions; (2) Proposal of a multiscale neighborhood technique for generating the initial decision maps and final decision maps. This design integrates the advantages of local neighborhoods with different sizes in an efficient manner, which improves the reliability of the focus measure; (3) Design of an averaging method based on decision maps at different scales to merge the borders between the focused and defocused regions, thereby suppressing the discontinuities between different focused regions in the fused results.