We present a method for comparing shapes of grayscale images in noisy circumstances. By establishing correspondences in a new image with a shape model, we can estimate a transformation between the new region and the model. Using a cost function for deviations from the model, we can rank resulting shape matches. We compare two separate distinct region detectors: Scale Saliency and difference of gaussians. We show that this method is successful in comparing images of fluid mixing under anisotropic geometric distortions and additive gaussian noise. Scale Saliency outperforms the difference of Gaussians in this context.