Registration of Images With Outliers Using Joint Saliency Map

Structure matching driven by joint-saliency-structure adaptive kernel regression

“…To avoid the above-mentioned mis-assignment and minimize the ourlier impact on the reconstruction of deformation field, we propose a robust weight mechanism by simultaneously considering the matching degree of local salient structures in the overlapping parts of the two images. In our previous work [31], it has been proved that the application of JSM is effective in tackling registration problems with outliers by emphatically grouping the JSSs into intensity-based similarity measure. Continuing the success of JSM, we further deploy the concept of JSM into our robust weight mechanism and pay more attention to the JSSs in the two images.…”

“…To solve the outlier problem, we proposed the joint saliency map (JSM) to group the corresponding saliency structures (called Joint Saliency Structures, JSSs) in intensity-based similarity measure computation [31]. The JSM has been proved to greatly improve the accuracy and robustness of rigid [31] [32] and nonrigid [10] [33][34] image registration with outliers.…”

“…To solve the outlier problem, we proposed the joint saliency map (JSM) to group the corresponding saliency structures (called Joint Saliency Structures, JSSs) in intensity-based similarity measure computation [31]. The JSM has been proved to greatly improve the accuracy and robustness of rigid [31] [32] and nonrigid [10] [33][34] image registration with outliers. We further think, by reflecting the local structure correspondence, the JSM also could guide the local kernel regression for accu-rately estimating registration transformations in the nonrigid image registration with outliers.…”

Application of Improved Genetic Algorithm in Practical Medical Image Registration

“…The theoretical basis of information theory is: The information reflected in the two images must have some intrinsic association, with the change in the matching degree of the two images, this correlation also changes. Mutual information can qualitatively be thought of as a measureof how well one image explains the other, and is maximized at the optimal alignment[4,5]. The maximization of mutual information[6], as a kind of information theory method for similarity metric, needs not to conduct segmenting or pretreatment to images nor limit treated contents, which has better precision and higher stability compared with other methods based on voxel similarity[7,8].…”

Image based representation for 3D content delivery

“…MI was used by Tsai and Lin [136] to devise the decision rules for transmission ordering. In [106], a joint saliency map-MI approach was proposed for the robust registration of images. Qin et al [106] uses the approach of joint saliency map to compare the saliency structures of two image to solve the problems of outliers and local maxima in MI-based image registration.…”