Ultrasonic image classification based on support vector machine with two independent component features

Imbalance of gender ratio at birth has been a serious phenomenon in China. To solve this problem, a scheme for ultrasonic image classification is proposed for preventing fetus gender examination with non‐medical purposes. Tens of thousands of ultrasonic images with and without sexual organs are collected to establish a professional database. These images are preprocessed firstly by cropping, de‐noising and compression. And then, independent component analysis (ICA) is applied for feature extraction under two architectures, which give local and global information respectively. The first architecture treats the images as random variables and the pixels as outcomes, while the second treats the pixels as random variables and the images as outcomes. After training of selected samples, a support vector machine (SVM) classifier which combined the two ICA representations is established for recognition, and a good performance is given for testing data. Finally, some new technique is suggested for algorithm improvement in the future.