Ultrasound image segmentation based on a multi-parameter Gabor filter and multiscale local level set method

To address the weakness and discontinuity of the edges and the uneven distribution of gray in ultrasonic images, an improved edge extraction algorithm based on a multi-parameter Gabor filter and multiscale local level set method is proposed. With the grayscale inhomogeneity of ultrasound images being regarded as texture in different directions, the directionalities of the Gabor wavelet are adopted to filter at different angles. An intermediate image is obtained to isolate the difference between each region and the background, which will allow the retention of the original image by maximizing it with a fusion method. The Gabor filter kernel with multi-center frequency meets the complex frequency distribution characteristics of ultrasound images, and the mean fusion method is used to maximize the information in the image while reducing noise influence. For the edge of the ultrasound image is weak and the grayscale is uneven, the local intensity clustering level set method is improved. A Gaussian convolution kernel template is applied with different variance sizes to fit the grayscale changes in different parts of the image. Testing the ultrasound images of a stomach show that correlation coefficient and sensitivity coefficient reaches 0.856 and 0.910, respectively, which is a 20.7% and 5% improvement over the traditional LIC algorithm, respectively. This method can satisfy the system requirements where non-contact, online, real-time, higher precision and rapid speed strong anti-jamming and stabilization are needed.