Improving Iterative Randomized Hough Transform for Automatic Detection of Fetal Head from Ultrasound Images

摘要

In this paper, we propose an improved iterative randomized Hough transform (IRHT) method to detect fetal head automatically in ultrasound images. With the update of region of interest (ROI) in the IRHT method, the noise pixels are gradually excluded from the region of interest during iteration process, and the estimation becomes progressively close to the target. In order to enhance the efficiency and stability of this algorithm, we consider introducing the number (N) of pixels on the detected ellipses, and select the ellipse with the maximal number of the pixels on the ellipse as the result for each iteration, which is selected from the top-M peaks in the accumulators of the whole detected ellipse samples. The experiments on fetal ultrasound images demonstrate that the proposed method achieves more robust and accurate results, and has a better performance for fetal head detection than the IRHT method.