| With the development of the imaging technology, images have become important tools for conveying, transferring and storing information. Given that a single sensor often cannot produce a complete representation of a scene, it is desirable to design fusion systems in such a fashion so as to enrich the visual experience of these users by combining images with complementary information. Image fusion can be used in medical diagnosis, camera manufacturing, remote sensing and object detection. Based on the characteristics of images, image fusion algorithms will be designed and studied in this paper, and the problems of objective image fusion evaluation metrics and their validation are also investigated. The main contents of this paper are summarized as follows:1. Medical structural and functional image fusion algorithmMedical image fusion technique has important application value in clinical diagnosis. For example, structural images can capture structural and anatomical information of human organs, and functional images can reflect the functional information. The image fusion can effectively improve neurological diseases judgment. But the structure of the two kinds of images is of great difference, and traditional algorithms are likely to result in color distortion and texture loss. This algorithm decomposes the structure image into high frequency layer and low frequency layer based on edge-preserving filter, and segments the functional image into the target and background area using the maximum between-class variance method. Then the target area and high frequency layer in structural image are combined to obtain the corresponding region of fused image, the rest regions of the fused image are filled by the original information of the structural image. Experimental results show that the proposed algorithm can effectively preserve the color and texture in fused image.2. Multi-focus image fusion algorithmsThe existing algorithms are faced with two main problems:(a) the fusion of smooth regions near edges is poor; (b) fused images are plagued by block effects. To solve the two problems, methods have been proposed in this paper.(1) Note that faster gray changes in smooth regions near edges do not necessarily means higher defition. Therefore, in our algorithm, special regions are identified by Sobel operator and distance transform, and the fusion strategy for them is designed to be opposite to the one for other regions. Experimental results show that the proposed algorithm improves the fusion quality of smooth regions near edges.(2) A new algorithm of image fusion based on spectrum comparison is proposed. The algorithm does not need block processing, but directly detects the focus of the whole image. Experimental results show that the algorithm can preserve clear information in source images and effectively avoid block effects. In addition, the fusion algorithm is also suitable for color image fusion.3. Multi-scale image fusion algorithmsTwo universal image fusion algorithms are proposed:(1) A new image fusion algorithm is proposed based on Internal Generative Mechanism (IGM). In the algorithm, source images are decomposed into a coarse layer and a detail layer by simulating the mechanism of human visual system perceiving images. The interests of the algorithm lie in the fact that it accords with the basic principles of human visual system perceiving images and it can preserve detail information that exists in source images.(2) A new image fusion algorithm is proposed based on simultaneous empirical wavelet transform. It contains three steps:image decomposition, coefficient combination, and image reconstruction. Compared with traditional ones, it can simultaneously decompose the source images into the same wavelet set which is the optimal one for these images.4. Image fusion performance evaluation measurePerformance evaluation is one of important research directions in image fusion field. In this part, two algorithms have been proposed.(l)The paper presented an effective metric set selection strategy. The metrics in the set should meet two conditions:(a) the evaluation accuracy is high; (b) the correlation between the measures is low.(2)The paper presents a new image fusion performance measure, which is composed of two parts:similarity measure and contrast measure. The former is used to measure the information similarities between original image and fused images; the latter is used to evaluate the contrast of the fused image.5. Validation on image fusion performance evaluation measuresObjective image fusion evaluation metrics play a vital role in choosing proper fusion algorithms and optimizing parameters in the field of image fusion.(1) In this paper, we proposed a novel validation method using ROC (Receiver Operating Characteristic) curve and AUC (the Area Under the ROC Curve). The proposed method takes the predicted quality scores into account rather than just counts how many fused images are correctly evaluated, which makes it more discriminating than other existing methods.(2) A new version of ROC curves is invented namely diAUC. Meanwhile, the area under diROC curves (diAUC) is adopted to measure the quality of classifiers quantitatively. The attractive feature of the proposed algorithm is that it takes classification difficulty of instance into account which makes the evaluation more precise and reliable.The major innovations of this paper include:(1) According to the characteristics of medical structural and functional images, an image fusion based on edge-preserving image decomposition and image segmentation algorithm is proposed. It can avoid color distortion and texture loss of fused images.(2) The problem of poor effect of smooth regions in multi-focus image fusion algorithms is addressed by detecting smooth regions and designing special fusion rules. A global spectrum comparison method is proposed to solve the problem of blocking artifacts.(3) The visual perception theory is brought into image fusion algorithm, source images are decomposed by simulating Internal Generative Mechanism, which guarantees that fused images are qualified for the perception mechanism of human visual system.(4) Simultaneous empirical wavelet transform is proposed, by which multi source images can be decomposed into one optimal wavelet space simultaneously. It is conducive to designing appropriate fusion rules.(5) The paper puts forward the theory of constructing objective evaluation measure set, and designs the basic method of construction.(6) The essence of the objective evaluation measure of image fusion is analyzed, and the performance of the measure is verified by the ROC analysis method from the aspect of the classifier. The concept of evaluation (classification) difficulty is introduced to improve the reliability of ROC analysis.
|
PDF Full Text Request