In order to overcome the disadvantages of color, shape and texture-based features definition for medical images, this paper defines a new kind of semantic feature and its extraction algorithm. We firstly use kernel de...In order to overcome the disadvantages of color, shape and texture-based features definition for medical images, this paper defines a new kind of semantic feature and its extraction algorithm. We firstly use kernel density estimation statistical model to describe the complicated medical image data, secondly, define some typical representative pixels of images as feature and finally, take hill-climbing strategy of Artificial Intelligence to extract those semantic features. Results of a content-based medial image retrieve system show that our semantic features have better distinguishing ability than those color, shape and texture-based features and can improve the ratios of recall and precision of this system smartly.展开更多
目的针对不同视点下具有视差的待拼接图像中,特征点筛选存在漏检率高和配准精度低的问题,提出了一种基于特征点平面相似性聚类的图像拼接算法。方法根据相同平面特征点符合同一变换的特点,计算特征点间的相似性度量,利用凝聚层次聚类把...目的针对不同视点下具有视差的待拼接图像中,特征点筛选存在漏检率高和配准精度低的问题,提出了一种基于特征点平面相似性聚类的图像拼接算法。方法根据相同平面特征点符合同一变换的特点,计算特征点间的相似性度量,利用凝聚层次聚类把特征点划分为不同平面,筛选误匹配点。将图像划分为相等大小的网格,利用特征点与网格平面信息计算每个特征点的权重,通过带权重线性变换计算网格的局部单应变换矩阵。最后利用多频率融合方法融合配准图像。结果在20个不同场景图像数据上进行特征点筛选比较实验,随机抽样一致性(random sample consensus, RANSAC)算法的平均误筛选个数为30,平均误匹配个数为8,而本文方法的平均误筛选个数为3,平均误匹配个数为2。对20个不同场景的多视角图像,本文方法与AutoStitch(automatic stitching)、APAP(as projective as possible)和AANAP(adaptive as-natural-as-possible)等3种算法进行了图像拼接比较实验,本文算法相比性能第2的算法,峰值信噪比(peak signal to noise ratio,PSNR)平均提高了8.7%,结构相似性(structural similarity,SSIM)平均提高了9.6%。结论由本文提出的基于特征点平面相似性聚类的图像拼接算法处理后的图像保留了更多的特征点,因此提高了配准精度,能够取得更好的拼接效果。展开更多
基金Supported by the National Natural Science Foun-dation of China(60572112) the Jiangsu High Education Natural Sci-ence Research Project (03KJD51002) the Fourth Group StudentResearch Project of Jiangsu University.
文摘In order to overcome the disadvantages of color, shape and texture-based features definition for medical images, this paper defines a new kind of semantic feature and its extraction algorithm. We firstly use kernel density estimation statistical model to describe the complicated medical image data, secondly, define some typical representative pixels of images as feature and finally, take hill-climbing strategy of Artificial Intelligence to extract those semantic features. Results of a content-based medial image retrieve system show that our semantic features have better distinguishing ability than those color, shape and texture-based features and can improve the ratios of recall and precision of this system smartly.
文摘目的针对不同视点下具有视差的待拼接图像中,特征点筛选存在漏检率高和配准精度低的问题,提出了一种基于特征点平面相似性聚类的图像拼接算法。方法根据相同平面特征点符合同一变换的特点,计算特征点间的相似性度量,利用凝聚层次聚类把特征点划分为不同平面,筛选误匹配点。将图像划分为相等大小的网格,利用特征点与网格平面信息计算每个特征点的权重,通过带权重线性变换计算网格的局部单应变换矩阵。最后利用多频率融合方法融合配准图像。结果在20个不同场景图像数据上进行特征点筛选比较实验,随机抽样一致性(random sample consensus, RANSAC)算法的平均误筛选个数为30,平均误匹配个数为8,而本文方法的平均误筛选个数为3,平均误匹配个数为2。对20个不同场景的多视角图像,本文方法与AutoStitch(automatic stitching)、APAP(as projective as possible)和AANAP(adaptive as-natural-as-possible)等3种算法进行了图像拼接比较实验,本文算法相比性能第2的算法,峰值信噪比(peak signal to noise ratio,PSNR)平均提高了8.7%,结构相似性(structural similarity,SSIM)平均提高了9.6%。结论由本文提出的基于特征点平面相似性聚类的图像拼接算法处理后的图像保留了更多的特征点,因此提高了配准精度,能够取得更好的拼接效果。
