基於逆遞迴式類神經網路之視訊色彩重現系統

Abstract

本論文提出一基於人類色彩感知模式之高畫質電視色彩校正機構。為達成 高傳真色彩重現之目的。此機構必需消除來自攝影機、終端機、彩色影像 媒體處理以及它們之間交互作用所產生之色彩誤差來源。其中，彩色影像 媒體處理可能包括有編碼╱解碼、調變╱解調及量化處理等。而輸入╱輸 出裝置之交互作用是攝影機與終端機間之色階與解析度轉換的不匹配。這 些色彩誤差源具有高度非線性的特質。為了消除色彩誤差源，我們提出一 經由逆遞迴式類神經網路控制器及正向模型所組構成的適應性控制機構以 便克服高畫質電視系統之非線性及模型辨認的困難。上述正向模型之功能 為將系統輸出誤差（色差）轉換為控制信號誤差以須訓練逆遞迴式類神經 網路控制器。再者，正向模型微分參數之取得乃藉由Hornik所提之多層類 神經網路模型辨認來決定。另外，色差的量測當於均勻色彩空間以量化。 最後，藉由實驗的進行將證明所提機構之有效性。 This paper presents a new adaptive color correction process for High Definition Television (HDTV) system based on human perception model. To achieve the high-fidelity color , it requires eliminating the color error sources resulted camera, monitor, intermediate color image processing and mutual interaction. The intermediate processing may include coding/ decoding, quantization, and modulation/demodulation. the mutual interaction denotes the color gamut mismatch and resolution conversion between camera and monitor. It can be that those color error sources of unknown form are highly nonlinear. Hence, a cost-effective adaptive control scheme consisting of both back-propagation neural net controller forward model is proposed to overcome the difficulty of with both nonlinearity and model-identification of the HDTV systems. The forward model is used to convert the plant error (color difference) into control signal error for the back-propagation controller. Furthermore, the values of derivative-related parameters of the forward model are by Hornik's multi-layer neural net identification process. In addition, it is noted that the measure of the color (plant output error) should be quantified by uniform color in order to perform the system control. Finally, the effectiveness of our method is being verified by a number of experiments.