Author(s)

Yanjun Chen, Xinke Li and Yuhong Zhang

Corresponding Author

Yuhong Zhang

ABSTRACT

A recovering method for the missing pixels without using any reference images is proposed to fill the gap filling for the Landsat-7 SLC-off images. The image is firstly preprocessed for the gap location, unsupervised classification and local average gray. Then, regression analysis is used to calculate the gray-level-change slope for the pixels on both sides of the stripe respectively in three directions of each missing pixel. Under the unsupervised classification criteria, we choose the filling direction of the current pixel of the gap by judging the classifications of the pixels on both sides of the gap in three directions. Next, different algorithms are used to fill the boundary pixels and the non- boundary pixels in the stripes. For the non-boundary pixels, cubic spline interpolation is used to calculate the gray value of the current pixel. Finally, all the stripes are filtered by an adaptive filter. The experimental results show that the recovery algorithm by the proposed method has better visual effect comparing with other three methods, and no stripe trace can be found in the unsupervised classification image, which lays a foundation for further research on geographic spatial analysis.

KEYWORDS

Interpolation, regression analysis, unsupervised classification, gap filling, no-reference-image recovering