摘要
本文针对BP神经网络对图像高频信息压缩效果欠佳和收敛性较差的问题,提出了将SPIHT(多级树集合分裂)算法应用于BP神经网络图像压缩处理的新结构。在研究中,首先对原始灰阶图像做小波变换;其次采用SPIHT算法对小波系数作量化编码处理;然后将产生的图像码流输入BP神经网络作进一步压缩处理。在实验中,本文提出的算法不仅在较大压缩比下仍能得到较高的峰值信噪比,而且有效地改善了图像的"块效应"问题;同时,提高了BP神经网络图像压缩的收敛速度,从而证明了该系统结构对图像压缩处理是有效的。
An image compression scheme for BP neural network with SPIHT algorithm is presented to solve the poor image quality and low convergence speed in high spatial frequency information compression of BP neural network image. In researching, the original grayscale image is converted to the wavelet coefficients using wavelet transform. SPIHT algorithm is adopted to quantize and encode the wavelet coefficients. And then the image code stream is transmitted to BP neural network for a further compression. In experiment, this system not only can get a better PSNR even under high compression ratio, but also avoid the ＂tiling＂ effect, and the convergence speed of BP neural network image processing is imoroved.
出处
《电子测量与仪器学报》
CSCD
2008年第6期7-11,共5页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金(编号:60429101)资助项目
