摘要： 在水印检测中，通常使用固定长度的样本，即检测时需要大量的待检测样本，这对于多水印检测和视频水印检测是不合适的。为此，研究连续水印检测，并设计改进方法。在对连续水印检测理论进行分析的基础上，发现操作特征函数指标及所需样本数量均与嵌入因子有关。该方法用局部神经网络对原图像进行估计，可以减小嵌入因子误差，提高连续水印检测性能。

关键词: 水印, 水印检测, 统计检测, 检测器, 检测性能, 局部神经网络

Abstract: In watermark detection, Fixed Sample Size(FSS) watermark detection needs large number of signal observations and it is not suitable for applications such as detecting multi-watermarks or video watermark detection. To overcome the difficulty, the sequential watermark detection is researched and an improved method is put forward. In analysis of the sequential watermark detection, the Operating Characteristic Function(OCF) and the Average Sample Number(ASN) are all related with the actual embedding factor. In order to improve the sequential watermark detector performance, a local network is applied to predict the original image because it can reduce the prediction error compared with the simple neighboring pix prediction, and improve the performance of sequential watermark detection.

Key words: watermark, watermark detection, statistics detection, detector, detection performance, local neural network

中图分类号: 

• TP391