Ground object spectral attribute-based dark fringe repair method

Abstract

The invention discloses a ground object spectral attribute-based dark fringe repair method. The method comprises the steps of firstly extracting a spectral curve corresponding to a non dark fringe part of a dark fringe pixel; based on a fragment of the spectral curve, searching for a ground object spectral curve with the highest similarity through a full image search; taking a ground object type corresponding to the spectral curve as a ground object type of the dark fringe pixel; and repairing a pixel value of the dark fringe pixel by using a corresponding spectral value. According to the method, a statistical relationship is no longer blindly utilized for repairing the dark fringe pixel, instead, priori knowledge is utilized for judging the ground object type of the dark fringe pixel, and a dark fringe pixel region is filled up by depending on spectral features of other same pixels imaged at the same time; and compared with a conventional method, the repair of the dark fringe pixel region is realized by depending on the spectral features of a ground object instead of blindly depending on statistical correlation, so that the repair method is scientific and reliable, and is higher in precision.

Description

technical field [0001] The invention relates to a method for repairing bad lines, in particular to a method for repairing bad lines based on spectral attributes of ground objects, and belongs to the field of image processing research. Background technique [0002] A bad line refers to an entire column or row of obvious dark lines on the image, and its pixel value is generally a small atypical pixel value, such as 0 or a very small value. Bad lines exist on many types of imagery, especially hyperspectral imagery. The main reason for its formation is the calibration problem of the sensor. Because it is very difficult to calibrate the tens of thousands of detection units of the area array CCD device. The existence of bad lines seriously affects the use of remote sensing images and must be corrected. Most of the existing bad line correction methods rely on the correlation between the bad line pixel and the left and right adjacent pixels, establish a statistical model, and use...

AI Technical Summary

Problems solved by technology

These methods have a strong dependence on adjacent pixels, lack of theoretical basis, and the repaired pixel value is an estimated value of statistical significance, rather than the real pixel value of the ground object.

Images