提出了NSA多尺度模型。该模型摒弃了LSA模型中不同尺度的图像间具有线性映射关系的假设。首先利用神经网络建立不同尺度图像间的映射关系；其次使用反向传播算法训练神经网络确定这种映射关系；最后根据该映射关系由低分辨率图像估计高分辨率图像。利用对比度相似性量化估计图像与目标图像间的相似程度。将该模型用于人脸识别，提出利用梯度算子进行图像分割提高识别的准确性。实验结果表明，以该模型分析得到的对比度相似性为95．3634%；以该模型为基础的人脸识别系统对光照具有很好的鲁棒性。

A multiscale mode named neural networks scale autoregressive(NSA)is presented. The model uses neural networks to represent the map between the pixels residing at images of various resolutions;in stead of the assumption of linearity in linear scale sutoregressive(LSA)model. Then back-propagation algorithm is used in the neural networks training to decide the map,finally,the identified map is used to estimate images at finer resolution from coarser versions. Similarity in contrast is used to gauge the degree of similarity between estimated images and target images. The model is applied in face recognition and image segmentation approach using gradient operator is proposed to increase the recognition accuracy. Experimental results show that similarity between image estimated by NSA model and target image is 95.3634%. The NSA-based face recognition system is robust to illumination.