Larynx Ultrasound Image Stitching Based on Multiconstraint Super-Pixel Feature

In order to obtain wide-field ultrasound images and comprehensively observe the larynx and its surrounding tissues, this article presents a larynx ultrasound image stitching algorithm based on multconstraint super-pixel features, which can achieve effective stitching of local images on both sides of the larynx. In this study, the 120 larynx ultrasound images from healthy volunteers of different ages were collected, including children, middle-aged, and old-aged. First, to perform the pre-aligning of local field images of larynx, it can be represented by homography matrix from Scale-invariant Feature Transform (SIFT) point and super-pixel region features. Then, a multiconstraint energy function was introduced to further adjust the registration, which include Gabor texture feature constraint and saliency perception. Finally, a linear weighting blending method was utilized to determine the optimal seam and finish stitching. The proposed method outperforms compared state-of-the-art methods in the visualization results in stitching larynx ultrasound images of three different age groups. Compared with the closest competitor, the peak signal-to-noise ratio (PSNR) improves 7.6%, 6.3%, and 5.8%, the structural similarity index (SSIM) improve 12.9%, 6.5%, and 14.3%. This demonstrated that the stitching image obtained by proposed method has the least distortion and the highest image quality. Moreover, the maximum absolute error does not exceed 0.251 cm, which can basically be consistent with the ground truth. These experiments results illustrate the effectiveness and robustness of the proposed method on larynx ultrasound image stitching, and provide an effective and feasibility solution for the ultrasound diagnosis of the larynx.