|
Elastic registration of biomedical images on CUDA-Supported graphics procesor units
Michálek, Jan ; Čapek, Martin ; Janáček, Jiří ; Kubínová, Lucie
Elastic registration is a task of finding the matching of two images, using geometric and elastic transformations, so that objects in images have the same size, position and orientation. We apply elastic registration in the framework of volume reconstruction, where an object acquired from parallel physical sections is composed and mutual positions of the sections including deformations caused by their cutting have to be found. The method lies in optimizing a functional consisting of two parts: first, discrete total variation as a measure of roughness and, second, L1 norm as a measure of dissimilarity of images. As a parallelizable optimization strategy we apply a potential-based equivalent transformation of a (max,+)-labelling problem. CUDA-based implementation of the described elastic registration algorithm is reasonably fast, requires seconds to minutes of calculations, provides good results and, thus, can be used for practical tasks dealing with alignment of biomedical images
|
| |
| |
|
Solving image analysis problems by minimization of total variation
Janáček, Jiří
A solution of various problems in image analysis using concurrent minimization of total variation and Lp loss function is presented. The minimization is achieved by a steepest descend method using graph cut minimization in each step. Regularization of noisy image and registration of microscopic images of physical sections are demonstrated
|
|
Ohodnocení chyby objemové rekonstrukce biologických vzorků z konfokálních obrazů
Čapek, Martin ; Janáček, Jiří ; Kubínová, Lucie ; Smrčka, P. ; Hána, K.
We performed both volume reconstructions using images captured by the USB microscope and images captured by the confocal microscope. We manually marked important corresponding structures in both reconstructed data sets, and computed distances between corresponding structures, assuming that structures in the reconstruction from USB microscope data are without deformations. According to our expectations, the main errors of high-resolution volume reconstruction (from confocal data) are in the direction of physical cutting (vary in units of millimeters) and in the direction perpendicular to cutting due to off-cut (vary in tenths of millimeters)
|