An Automatic Mosaic Method Of Large Field View And Multi-Channel Remote Sensing Images Of TDICCD Cameras

In the field of Image mosaic, the current main mosaic methods include: feature extraction -based image registration, including the corner and contour extraction; phase correlation-based registration in frequency domain; registration based on pixel intensity. Feature extraction method has fast speed, but its accuracy depends directly on the accuracy of feature extraction. Phase correlation-based method has certain anti-noise and anti-jamming, but its application is relatively limited by the number of overlapping pixels. Intensity-based method is simple and intuitive, but has relatively poor immunity.TDI (time delay and integration) is the scanning technology that can increase the sensitivity of line scan sensors. Large field view TDICCD camera is butted by multi-chip TDICCD and outputs remote sensing images by multi-channel, so it mainly has the following characteristics in space imaging: 1) Different line frequency. 2) Different collection time on the same scene. 3) The impact of TDICCD spatial orientation. 4) The flutter generated by momentum wheel, jet, point control and the solar panels. 5) The impact of weather and terrain conditions.Because of the complexity of the above factors, In terms of mosaic of large field view and multi-channel TDICCD remote sensing images, noise immunity and robustness of the previous classical method is not high, which is difficult to achieve high accuracy and speed image mosaic. For that, this paper proposes a spatial registration method which is based on correlation coefficient function and find out match points by self-test search method, in order to reflect the overall and partial registration parameters and achieve accurate registration. First, because of the characteristics of overlapping pixels between multi-channel TDICCD images, the method uses the cross correlation of variable search window as the evaluation function; Secondly, the local detection method is employed to monitor change of parameter to figure out the best parameters of each part of images; Besides, error matching points are controlled by interference elimination method, so that parameters is more credible; Finally, sub-pixel locating algorithm is proposed to reduce error of image registration.Multi-channel images taken by XXX satellite are mosaicked for comparing the proposed algorithm and existing algorithms. The results of practical remote sensing image mosaic show: accuracy of the algorithm excess 0.1pixel, while rapidity, stability, noise immunity and robustness are higher than other methods and image mosaic obtains expected results.