针对多聚焦图像融合过程中可能出现的问题,如细节丢失、边缘伪影和区块效应等,提出了一种基于纹理特征提取的多聚焦图像融合方法.通过纹理特征提取算法获取图像纹理细节,用引导滤波对细节进行增强和细化处理.对滤波后特征图采用像素极大值策略生成初始决策图,利用小区域去噪方法得到准确的决策图.结合原始图像与最终决策图进行图像融合,生成全聚焦图像.结果表明,该方法融合的图像质量有显著提升,与其他方法相比,均方误差评价值降低了约37.14%,同时,基于归一化互信息度量、基于Tsallis熵度量和基于非线性相关信息熵度量的评价参数值分别提升了约26.1%、6.18%和13.2%.此外,该方法还可以充分保留图像细节信息且没有区块效应和边缘模糊的现象,具有较强的实际应用价值.
To address the challenges associated with multi-focus image fusion,such as detail loss,edge artifacts,and block effects,we proposed a texture feature extraction-based method for multi-focus image fusion.The texture details were extracted from the image using a texture feature extraction algorithm,followed by enhancement and refinement through guided filtering.The initial decision map was first generated using a pixel maximum strategy on a filtered feature image,and then was refined to get a final decision map using a small region denoising method.Finally,a fully focused image was obtained by fusing the original image and the final decision map.The results show this method significantly improves the quality of the fused images.Compared with other methods,it reduces the mean square error evaluation value by about 37.14%,meanwhile the evaluation parameter values based on normalized mutual information metric,Tsallis entropy metric,and nonlinear correlation information entropy metric increase by about 26.1%,6.18%,and 13.2%,respectively.In addition,the method fully preserves image detail information without block effects or edge blur,which shows its significance in practical applications.
彭聪;李万乔;李立仁;焦永鑫;刘晓华
武汉工程大学机电工程学院,湖北 武汉 430205武汉工程大学计算机科学与工程学院,湖北 武汉 430205
计算机与自动化
多聚焦图像融合;纹理特征提取;特征增强;引导滤波
multi-focus image fusion;textural features extraction;feature enhancement;guided filter
《武汉工程大学学报》 2024 (002)
197-202 / 6
国家自然科学基金(62171329)
10.19843/j.cnki.CN42-1779/TQ.202308011
评论