多任务学习框架下激光全息图像超分辨率重建

Super resolution reconstruction of laser holographic image based on multi task learning framework

传统激光全息图像重建方法需要训练样本多,计算复杂度高,导致图像配准误差,为了解决该难题,提出了多任务学习框架下的激光全息图像超分辨率重建方法。首先对激光全息图像进行退化处理,然后采用不同方向平移量及旋转角度参数对激光全息图像进行精度配准,最后根据非局部均值构建超分辨率重建模型,实现多任务学习框架下全息图像超分辨率的重建。结果表明,该方法能够精确地实现低像素激光全息图像的超分辨率重建,有效地控制了误差水平,提高图像的清晰度。

The traditional method of super resolution reconstruction of laser holographic image, need a large amount of training set, large amount of computation, poor performance in convergence and accuracy, image registration has a great error. A multi task learning under the framework of the laser holographic image super-resolution reconstruction method presented in this paper, the first treatment processing of laser holographic image degradation analysis, approximation to estimate the degradation function and introduce noise, laser holographic image restoration processing; introduced in different directions of translation and rotation parameters for laser holographic image do sub-pixel accuracy of registration; finally construction of super resolution reconstruction model based on non local means. To solve the super resolution function, finally realizes the multi task learning under the framework of the holographic image super resolution reconstruction. Experiments show that the proposed method can accurately complete the super-resolution reconstruction of low pixel images, and can effectively control the level of error, and improve the resolution of the image.