多特征融合的图像自动变形
Multi-feature fusion-based image morphing
- 2014年19卷第7期 页码:1012-1020
网络出版:2014-07-01,
纸质出版:2014
DOI: 10.11834/jig.20140704
移动端阅览

浏览全部资源
扫码关注微信
网络出版:2014-07-01,
纸质出版:2014
移动端阅览
图像变形算法中特征基元提取和匹配方式大部分都是采用人机交互的方式进行,并且在遮挡区域变形时出现较多的鬼影和模糊现象,使得针对同一场景图像变形实现繁琐且效果不佳,针对这些问题提出一种基于多特征融合的自动图像变形算法。 该算法提取多种图像特征信息(如Surf特征算子、Harris算子、Canny算子等)并进行多特征融合匹配,得到一个分布适当且对应关系正确的三角网格,再结合图像变形,实现自动图像插值。 实验结果显示,自动的提取特征基元有效地减少了人工操作,而多特征融合匹配有效地抑制了图像变形时边缘或遮挡区域鬼影的产生。 提出的融合匹配方法,将不同的特征信息有效地融合匹配从而改善了图像变形算法。通过对多组实验结果进行问卷调查,91%的参与者认为该算法有效地改进图像变形结果。
Image morphing algorithm is a branch of image-based rendering(IBR). Normally
it extracts features and matches features by human-computer interaction. However
there are problems in such human-computer interaction algorithms.When occlusion areas are processed
ghosting and blurring have a great chance to occur which are fatal to image morphing algorithms. All these phenomena lead to poor experimental results in the same scene. The implementations of old-fashioned image morphing algorithms are always complicated and inefficient and usually not suitable for practical application. In order to solve these problems
we propose a novel and efficient image morphing algorithm based on multi-feature fusion in this paper. In spite of marking two relevant images on edges
corners
and rich-texture areas by human-computer interaction
our proposal innovatively extracts multiple-image feature information (such as Surf feature
Harris feature
and Canny feature) with multi-feature fusion matching
which obtains a properly distributed triangle mesh pair with correct correspondence. Then
automatic image interpolation is achieved by the conjunction of triangle mesh and image morphing. We select Surf feature
Harris feature
and Canny feature as basic features. First
we extract these image features from the original image and destination image. Then
we process Surf feature based on Delaunay triangulation to obtain an initial triangle mesh. It is fusion matched with Harris features or other image features. According to this step
an accurate and uniform triangle mesh is acquired. We also define a matching cost function
a feature point of color intensity cost
and a grade matching cost to optimize the matching of features. It improves the accuracy of image feature matching. Finally
the image is transformed based on the acquired triangle mesh to a virtual view image between the original image and destination image. In the conventional image morphing algorithm
the system will take a long time to choose feature points manually which is not suitable for practical application. In our proposal
the step of extracting image features and matching image features is achieved automatically so that it compensates the inefficiency of traditional image morphing algorithms. Compared with the method of human-computer interaction
the performance of our proposal is extremely better in details such as edges
corners
and rich-texture areas. Theoretic analysis and experimental results indicate that automatically extracted and matched image feature method effectively reduces the manual operation
and multi-feature fusion effectively restricts the ghosting produced at edges and occlusion areas in image morphing. Image morphing algorithm is widely used in domains such as special effects in movis and 3D TV. However
the complicated and excessive implementation of traditional image morphing algorithms restricts the range of application. In this paper
a new approach of fusion matching is proposed. During this approach
different types of image features are effectively matched to improve image morphing algorithm. Through the questionnaire survey to the experimental results
91% of the participants think that the algorithm effectively improves the image morphing result. With the proposed approach
image morphing algorithm has great potential in the application of other domains.
相关作者
相关机构
京公网安备11010802024621