Research On The Compressive Sensing Based Multiple-focus Image Fusion Techniques

As an effective technical solution, image fusion can improve image availability, definition, identification and can be a comprehensive analysis of scene and target by effective use of complementary and redundant information. Compressive sensing theory makes a breakthrough to Shannon sampling theorem on the limit of the sensor system. As long as the signal is sparse with a base, it can project the high dimentional signal to a low dimension space, via a measure matrix which not related to the base, directly get the reversible and compressive signal. And don’t need high-speed sampling rate of the original signal. Compressive sensing can help to reduce the need for storage space and computing cost of image fusion in large data.Research on the pixel-level image fusion method based on compressive sensing, this paper propose a new adaptive sampling mode, introduce saliency measure and harmonic coefficients, and introduce an improved fusion rules based on global and high-low saliency measure and harmonic coefficients. This paper analyzes the influence of different sampling modes, different fusion rules, different parameters on different image fusion performance. The proposed and improved scheme are simulated and compared with the existing CS fusion scheme in different images, respectively. And they are analyzed and verified in the application of extended depth of field, multi-exposure and multi-angle light source image.Firstly, this paper studies the fusion performance of the adaptive sampling mode, compared with the existing CS sampling mode. Experimental results show that, adaptive sampling mode can adaptively add more sampling points in the area where the frequency changes significantly, and therefore grasp the edge of the image and texture direction better. For the images whose DFT spectrum has significant changes, in the case of a relatively low compression rate, the fusion effect of the adaptive sampling mode is better than that of the existing CS mode. And the performance of CS image fusion scheme based on adaptive sampling mode is not sensitive to the scale parameter (i.e. no major changes with the scale parameter).This paper analyzes the impact the choice of weighted factor on image fusion performance. Due to the choose of fusion weighted factor which can reflect the characteristics of images has a direct impact on image fusion quality. Experimental results show that, four different saliency measure (amplitude, energy, variance, mean variance) can all reflect the characteristics of images in the globai and high-low mode, and have a similar performance. And the performance of CS image fusion scheme based on high-low frequency saliency measure is not sensitive to the radius parameter. In addition to the image whose average luminance difference is large, the improved harmonic coefficient can also reflect the important feature of original image, and have a similar performance with the fusion scheme based on global and high-low saliency measure. With the increase of frequency measurement, the performance of CS fusion scheme increases. And the images whose spectrum is more sparse, can get accurate result via less measurement(more large compression rate).The proposed scheme and the existing CS fusion scheme are simulated and compared with different images. The experimental results show that the proposed scheme has a better performance. In the scenes of typical extended depth of field, multi-exposure and multi-angle light source image, the CS scheme proposed in this paper was simulated and compared with the traditional wavelet and NSCT Fusion method. Experimental results show that, compared with the traditional method, the CS scheme can have a good result with less measurement. Especially for multi-angle light source image whose DFT spectrum has significant changes, the vision effect of the fusion image got by the improved CS fusion method is better than the traditional wavelet and NSCT method. Because harmonic coefficient can’t have a good measure of the important information proportion for images with large average luminance difference, the improved image fusion scheme based on harmonic coefficient is not fit for multi-exposure image. For different images, the objective and subjective evaluation are may not one-to-one corresponded. Generally for different applications and requirements, the objective and subjective evaluation are carried out jointly in research.