摘要
针对图像滤波的可硬件化和细节保留中存在的问题,提出一种多目标学习的图像滤波电路函数级进化设计方法.首先建立平均绝对误差最小和高误差点数量最少的多目标滤波学习模型,以函数级基因表达式为进化电路个体的表征;采用离线进化模式搜索近似目标的最优进化电路个体,并对该个体的基因表达式进行VHDL转换,再将转换的VHDL移植于可编程逻辑器件上实现滤波电路.与多种滤波方法在边缘保留、峰值信噪比-均方误差的比较结果表明,电路在细节和边缘保留上有较大的提高,视觉效果更好.
In order to solve the problem of filter hardware and reservation details of the image, this paper proposes a multi-objective evolutionary approach for image filter at function-level. Firstly, the multi-objective optimization model for image filter to minimize the number of large noise pixels and mean absolute error is constructed, and then, gene expression is used as the representation of the image filter, and a two-step off-line search method is to evolve a near-optimal image filter. Finally, the best-evolved gene expression is translated to VHDL and the VHDL is load onto programmable logic device for implementing the image filter. In experiments, three different types of noise are loaded on various images for filter training. The evolved filter by our approach compares various filters, in terms of peak signal to noise ratio （PSNR）, mean squared error （MSE） and the effect of filtering. The experimental results illustrated that the proposed multi-objective evolutionary filter is superior to previous filters.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2014年第9期1487-1493,共7页
Journal of Computer-Aided Design & Computer Graphics
基金
国家青年科学基金(61203218)
江苏省高校自然科学研究面上项目(13KJB520023)
苏州大学青年教师科研预研基金(SDY2013A16)
关键词
图像滤波电路
进化电路
峰值信噪比
函数级进化
多目标进化
基因表达式
image filter
evolutionary circuit
peak signal-to-noise ratio
function-level evolution
multi-objective evolution
gene expression
