A Destriping Method In Transform Domain For Line Scan Image

Stripe noise in line-scan image is a unique type of noise. It affects the visual effect of theimage seriously. It is easy to be recognized as linear object by mistake when performingimage analysis. So it has high theoretical value and practical significance to study thedestriping method. In transform domain, the noise is easy to be removed effectively whilekeeping more details untouched. According to the characteristic of the stripe noise, adestriping method based on wavelet decomposition and Fourier transform was proposed inthis dissertation. Concrete research content is as follows.1. The common destriping methods at present were analyzed. The two typical destripingmethods in transform domain, i.e., Fourier transform based destriping and waveletdecomposition based destriping, were analyzed emphatically.2. The causes of the stripe noise in line-scan color image were studied. Ideal model andpractical model of the stripe noise were established according to the distribution characteristicof the stripe noise.3. A de-striping approach based on combined wavelet decomposition and Fouriertransform is proposed. Firstly, the image that contains stripe noise was decomposed bymulti-scale wavelet to separate the vertical stripe noise information out of the image. Thisconcentrated the vertical stripe noise into the vertical details sub-band. Then the verticaldetails sub-band was transformed by fast Fourier transform to compact the stripe noiseinformation into a horizontal narrow area. The stripe noise was eliminated effectively afterperforming a Gaussian filtering on the Fourier transform coefficients. At last, the destripedimage was obtained after Fourier reverse transformation and wavelet reconstruction.4. Simulation experiments were performed using the line-scan color images of foreignfiber collected in practice. The experimental results show that the proposed method caneliminate the striping noise effectively while preserving more details of the original image.