ID 原文 译文
41226 针对超大视场红外图像畸变大、与人眼视觉差异明显的问题，提出了一种基于精确模型和逆投影的超大视场红外图像畸变校正算法以改善其视觉效果。 In order to solve the problems of large distortion of ultra-wide field of view(FOV)infrared image and obvious difference from human vision, in this paper, an ultra-wide FOV infrared image distortion correction algorithm is proposed based on accurate model and back-projection.
41227 该算法首先利用精确模型对超大视场红外相机成像中的物、像关系进行描述; The algorithm first describes the relationship between object point and image point in the ultra-wide FOV infrared camera imaging by using the accurate model.
41228 然后，针对红外图像像素采样率不高的缺点，利用较为精确的三次卷积插值法对图像进行插值来补全成像信息; Then, aiming at the disadvantage of low sampling rate of infrared image pixels, the more accurate cubic convolution interpolation method is used to interpolate the image for completing the imaging information.
41229 最后，根据校正图像上的待赋值像点的坐标，结合校正模型和超大视场红外相机精确模型，计算该像点逆投影到插值图像时的对应坐标， Finally, according to the point coordinates of corrected image, combined with the correction model and the accurate model, the corresponding coordinates on the ultra-wide FOV infrared image point coordinates are calculated by backward projection.
41230 并以最近邻像点像素值作为校正后图像像点的赋值。 The pixel value of the nearest image point is used as the value of the corrected image point.
41231 车载道路场景下的超大视场红外图像畸变校正实验结果显示，所提出的算法图像校正结果边界清晰、无锯齿效应，对场景中的直线平均还原偏差小于0.35pixels，表明该算法对超大视场红外图像畸变校正具有较好的适用性。 The experimental results show that the proposed algorithm has clear boundary and no sawtooth effect, and the average restoration error of the straight line in the scene is less than 0.35 pixels, which indicates that the algorithm has excellent applicability for the distortion correction of the ultra-wide FOV infrared image.
41232 针对现有紫外成像仪中紫外光与可见光图像配准实时性差，精度不高等问题，提出一种基于卷积神经网络(Convolutional Neural Networks，CNN)与小波融合(Wavelet Fusion，WF)的紫外光与可见光图像配准融合方法，并将其应用于高灵敏紫外成像仪中。 Aiming at the problems of poor real-time performance and low accuracy of the existing UV imagers in UV and visible image registration, a method of UV and visible image registration fusion based on convolutional neural network(CNN)and wavelet fusion(WF)is proposed and applied to high sensitive UV imager.
41233 首先，结合刚体变换和卷积神经网络对采集到的图像数据进行参数模型预训练，通过自主挖掘图像特征寻找到最优空间变换参数，实现紫外光图像与可见光图像的精确配准; Firstly, the parameter model of the collected image data is pre-trained by combining the rigid body transformation and convolution neural network, and the optimal spatial transformation parameters are found by self-mining image features to achieve accurate registration of UV image and visible image.
41234 其次，利用二维小波分解与重构算法实现紫外光与可见光图像的融合。 Secondly, twodimensional wavelet decomposition and reconstruction algorithm is used to realize the fusion of UV and visible images.
41235 实验结果表明，所提方法的紫外光图像与可见光图像配准速度快，叠加精度高，且具有良好的稳定性。 Experimental results show that the proposed method has fast registration speed, high overlay accuracy and good stability.