Jawad F. Al-Asad
Discrete Cosine Transform, Discrete Fourier Transform, Discrete Hartley Transform, Walsh-Hadamard Transform, Wavelet Transform
This paper presents new approaches to filter out multiplicative noise from ultrasound images. In the first approach, the speckle-noisy image is segmented into small dyadic-length segments. The 1D Discrete Cosine transform (DCT), Discrete Fourier transform (DFT), Discrete Hartley transform (DHT) and Walsh-Hadamard transform (WHT) are then independently applied on each of these segments. To filter out the noise, the corresponding inverse transform is applied on a subset of large magnitude coefficients. Also presented is a modified Wavelet de-speckling technique to improve performance by convolving the speckle-noisy image with a mask designed to drill speckle noise clusters. The drilled image is then segmented into small dyadic-length segments. To filter out the noise, a 1D wavelet scheme is used to de-noise each small segment independently. When applied on simulated and real ultrasound images, these approaches have outperformed popular nonlinear de-noising techniques such as 2D Wavelets and 2D Total Variation. They also showed less sensitivity to outliers resulting from the log transformation of the multiplicative noise.
Important Links:
Go Back
