Abstract
In this paper, we introduce a new self-adaptive masking method for shape detection of low Signal to Noise Ratio (SNR) images to improve the tracking capabilities of the motion correction in nuclear cardiac imaging. The method is developed using two-dimensional fast Fourier transform, ideal filtering in the frequency domain, recursive thresholding, and region recognition. This method is independent of the correlation between the contexts of the planar images and has a good tolerance for low SNR images. In addition, it is robust under the circumstances of significant abrupt motion of the object.