Computer Engineering and Applications ›› 2013, Vol. 49 ›› Issue (17): 159-163.

Previous Articles     Next Articles

Image local content amplification method based on SC-HOG detection

SUN  Yinghao, TANG Di   

  1. College of Computer and Information Technology, Liaoning Normal University, Dalian, Liaoning 116081, China
  • Online:2013-09-01 Published:2013-09-13

基于SC-HOG目标检测的图像局部内容放大方法

孙英皓,唐  棣   

  1. 辽宁师范大学 计算机与信息技术学院,辽宁 大连 116081

Abstract: Target-detective content amplification method can make different attentions to the content of photograph, and meanwhile maintain the integrity of the image. When detecting the target region, in the image operating process, it can maintain the regional content consistency, and make the content of the region have obvious amplifying effect. It uses the detection algorithm based on the Histograms of Oriented Gradients(HOG) to extract image characteristics information in multi-scale space, puts the HOG descriptor into linear Support Vector Machine(SVM) which as a classifier for recognition, uses the non-maximum suppression algorithm to fuse these bounding boxes to get the best bounding box of the target object, then uses weights function(M)to change the energy of each pixel in the area of bounding box, removes some seams which have less correlation with the surrounding area, lets the regional content reach more reasonable amplification effect.

Key words: seam carving, histograms of oriented gradients, support vector machine

摘要: 目标检测式内容放大方法可以对图片的内容采取不同的关注度,并且能够维持图像的整体效果。当检测出目标区域后,在对图像进行操作时,能够保持该区域内容的连贯性,使该区域的内容有明显的放大效果。利用基于方向梯度直方图(HOG)的目标检测算法在多尺度空间中提取图像的特征信息,并将所得的HOG描述符通过线性支持向量机(SVM)进行识别分类后,利用非最大化抑制算法对所得的多个边界框做融合处理得到包含目标物体的最优边界框,再用权值函数M改变边界框相关区域内各像素点的能量值,通过移除一些与周围区域相关度小的接缝,从而使目标区域内容达到更合理的放大效果。

关键词: 接缝雕刻, 梯度方向直方图, 支持向量机