Search - feature based o

Search - feature based o - List

[File Format] Novel-approach-for-texture DL : 0: 为提高基于内容的图像检索系统中纹理特征提取的有效性,提出了又一种纹理图像检索方法。该方法利用非下采样 Contourlet变换对图像进行分解, 提取不同子带和不同方向变换系数矩阵的均值和方差为特征向量, 作为数据库中纹理图像的索引,并利用两种不同的相似度函数计算图像之间的相似度,建立了一套基于示例查询图像的纹理图像检索系统。实验结果表明,与小波包等特征提取方法相比, 该方法不仅能降低特征向量维数,而且能取得更高的检索准确率和检索速度。-To i ncrease t he vali d ity o f tex t ure feature ex tracti on in conten t-based i mage retrieva l syste m, a nove l approach for tex ture i m age retr ieva lw as proposed . Th i s approach w as based on theNonSubsamp l ed Con t our l et T ransform ( NSCT). The m eans and variab l es of NSCT co efficien tsm a trix i n d ifferen t s ubbands and var i ous directi ons were ex tracted to for m the feature vectors wh ich we re reg arded as i ndexes of tex t ure i mages i n i m age da tabase . Two s i m il ar ity functi ons were used to compute the si m ilar i ty bet w een i m ages . A tex ture re trieval sy stem based on query i m age w as deve l oped . Co m pared to the w ave let packag e transform, th i s approach can no t on l y reduce the di m ension o f feature vectors , but a l so get higher accu racy and speed of retr i eva. l
Date : 2025-12-26 Size : 338kb User : jjdjjf
[File Format] Texture-Segmen-ta-t-ion-withWavelet DL : 0: 为了提高纹理图象分割的边缘准确性和区域一致性以及降低分割错误率, 提出了一种基于小波变换的利用特征加权来进行纹理分割的方法. 该方法包括特征提取、预分割和后分割 3 个阶段, 其中, 特征提取在金字塔结构小波变换的基础上进行预分割利用均值聚类算法来对原始图象进行初步的分割后分割则根据预分割的结果对特征进行加权, 然后利用最小距离分类器来实现图象的最后分割. 与传统的方法相比, 该方法在分割错误率、边缘准确性以及区域一致性等方面均有明显的改善-To imp rove the accuracy of boundary locat ions and region homogeneity as w ell as to reduce the er ro r rate in texture image segmentat ion, a novel app roach based on w avelet2t ransfo rm and using feature w eigh t ing is p ropo sed in th is paper . Th is new technique contains th ree consecut ive stages: feature ext ract ion, p re2segmentat ion and po st2segmentat ion . In the feature ext ract ion stage, texture features are ext racted by using the pyram id2st ruc2 tured w avelet t ransfo rm. The o r iginal image is then segmented init ially using themeans cluster ing algo r ithm in the p re2segmentat ion stage . A cco rding to the p re2segmentat ion results, the ext racted features are w eigh ted and the p re2segmented image is fur ther p rocessed w ith a m inimum distance classif ier in the po st2segmentat ion stage to f i2 nally get the segmented image . A ll technical po ints are clear ly descr ibed and p resented in detail . Some segmenta2
Date : 2025-12-26 Size : 151kb User : jjdjjf