【作者】 许晓诣；

【导师】 曾孝平； 田逢春；

【作者基本信息】 重庆大学 ， 信号与信息处理， 2007， 硕士

【摘要】 光学小波变换是近年来发展起来的实时处理信号的一种方法,它结合了小波变换和光信息处理的优点,以并行性和高速实时性为特点,为图像数据压缩提供了一个有效途径。这在需要大量图像数据传输的系统,如太空遥感、森林防火等领域有很大的应用前景。光学小波变换在模式识别和纹理分割方面已取得一定的成果,而对于用光学小波变换实现图像数据压缩的报道很少,其主要原因是它对整个光学变换系统的精度要求比其它应用高得多。在典型的光学4f系统的谱面上加滤波片是实现光学小波变换的常用方法。可以用光栅、空间光调制器、计算全息、胶片作为滤波片。论文研究了在4f系统的谱面上加胶片作为滤波片来实现小波变换的方法。主要工作如下:(1)研究了胶片的制作:分析了激光打印机、热升华打印机、胶片记录仪的优缺点,结合课题要求,确定胶片记录仪作为胶片制作仪器。(2)分析了胶片的误差:设计了一系列的样本图像作为色标,通过扫描仪和Matlab对胶片制作的样本图像进行定量定性的分析,得到不同颜色通道的误差特性曲线。(3)通过多项式拟合,寻求胶片图像均值与标准图像均值间的函数f(x),使f(x)与标准的数据点最接近。(4)胶片一旦制作,难以修改,针对胶片的加工工艺误差,利用神经网络自学习特性对胶片制作误差进行分析,用胶片制作误差修正神经网络权值。若用网络输出结果对标准图像进行预畸变处理,则制作后胶片图像的灰度值趋进标准图像的灰度值。研究结果表明:以24bit的彩色图像为分析基础,胶片图像的灰度级与原始图像的灰度级之间并非线性对应,R、G、B三通道的精度略有差异,R通道的性能最优,B通道的性能最差,当各通道灰度级以相同规律变化时,引入的误差最小。经多项式拟合拟合后,均值的平均峰值信噪比可提高10~20dB。而文中确定的网络模型能够稳定地收敛于较高的精度,为后续的误差补偿提供了依据。更多还原

【Abstract】 Optical wavelet transform is a real-time signal processing method which has developed in recent years. It combines the advantage of both wavelet transform and optical information. It is an effective approach to realize image data compression with the characteristic of parallelism and high-speed. It plays a significant role in some application fields which need to transmit abundant image data; for example, sunspace remote sensing and forest fireproofing. By now, it has got achievement on pattern recognition and texture segmentation. Because image data compression makes special demands on precision of the whole optical system, the reports on this field are rare. Placing a filter on the Fourier plane in an optical 4f system is a typical way to implement optical wavelet transform. Grating, spatial light modulator, computer- generated hologram and film can be used as the filter. This thesis focused on the research of the filter made of film. The whole thesis mainly consists of the following contents:Comparing with laser printer and dye-sublimation printer, film recorder of converting digital image into film was determined judging by the need of this project. Considering inaccuracy brought in by man-made factors or hardware devices in the facture of optical filter, it is necessary to measure the error of it. A feasible experiment scheme for measuring the error was provided.For the error, it is impractical to compensate after fabrication. In order to get relative accurate gray level on films, this paper borrowed ideas from predistortion applied to modify original image according to the nonlinear effects in manufacturing process. Therefore, a primary task of the paper is to set up the error model through artificial neural network, and then compensate the error in the system.It is verified by lots of experiments that based on color images of 24 bits. The relation between original image gray level and the one made by film is non-linear, and gray levels of film image varies with different grey level combination of R、G、B channel. Meanwhile, the precision of each channel is different. The performance of R channel is the best, and B channels’is the worst. By polynomial curve fitting, the PSNR of film will increase 10~20dB.Finally, the author made a conclusion and pointed out the problems that need to be explored further and the future research emphasis.更多还原