|
Infrared focal plane linear array detector is the core element to achieve high spatial resolution and high radiometric resolution infrared imaging. Due to the difference of detector response rate caused by non-uniform materials and the high sensitivity of camera to background signal changes, the complex non-uniformity of the image is eventually caused. In order to identify targets effectively and quickly, real-time non-uniformity correction must be performed on infrared camera images. The most commonly used method is to obtain the correction coefficients of each pixel based on variable temperature blackbody calibration before each image. However, it is difficult and costly to design full-aperture calibration blackbody for infrared camera with wide field of view and high resolution, and the feasibility is not high. In order to correct the non-uniformity of the acquired image, this paper proposes a non-uniformity correction algorithm for single-scene infrared image based on finite constraints. According to the principle that the pixel response rate is basically unchanged, the algorithm establishes a constraint relationship between the response rate of each probe element obtained by laboratory calibration and the response rate of the central standard pixel to iteratively screen the samples involved in the correction coefficient calculation, so as to realize the non-uniformity correction of the image. The proposed algorithm was verified by using the location imaging data of the medium-wave infrared camera. The non-uniformity of local uniform image was 44.1% before correction, and decreased to 4.4% after correction, while the uniformity of conventional scene correction was 13.5% after correction. Therefore, the non-uniformity of the proposed algorithm is reduced by 9.13% compared with the conventional scene correction method. The algorithm successfully breaks through the design limitations of no star calibration device of the wide-field high-resolution infrared camera, improves the target recognition efficiency of infrared image, and is conducive to implement in engineer. In addition, it only takes about 1min to correct an image of 30000 pixel *30000 lines by testing.
|
