CAJ | 학술논문

相关反馈方法越来越多地应用到图像检索领域，而现有的相关反馈方法鲁棒性较差，由此提出一种新的相关反馈算法。该算法首先对查询图像进行颜色、纹理、形状的特征提取，组成特征向量；然后进行特征相似比较，返回初始检索结果，在反馈阶段，通过标记的相关和不相关图像构造用户反馈矢量作为查询偏好，用已经建立好的图像之间的资源分配矩阵乘以用户反馈矢量，实现资源的扩散和图像重排；最后把重排后的图像返回给用户，并提出一种改进资源分配矩阵的方法。实验结果表明，算法不仅鲁棒性强，而且反馈效果出色。
상관반궤방법월래월다지응용도도상검색영역，이현유적상관반궤방법로봉성교차，유차제출일충신적상관반궤산법。해산법수선대사순도상진행안색、문리、형상적특정제취，조성특정향량；연후진행특정상사비교，반회초시검색결과，재반궤계단，통과표기적상관화불상관도상구조용호반궤시량작위사순편호，용이경건립호적도상지간적자원분배구진승이용호반궤시량，실현자원적확산화도상중배；최후파중배후적도상반회급용호，병제출일충개진자원분배구진적방법。실험결과표명，산법불부로봉성강，이차반궤효과출색。
Relevance feedback methods are increasingly applied to image retrieval field .For existing relevance feedback methods lack enough strong robustness , we present a new relevance feedback algorithm .First of all, the colour, texture and shape features of the query image are extracted respectively to form the feature vector , then the feature similarity comparison is executed , and is returned to initial search results;in feedback stage , the user feedback vector is constructed by marking the relevant and irrelevant images and is used as the query preference, user feedback vector resource is multiplied by the resource allocation matrix between images which has been established , and the proliferation of resources and the re-ranking of images are achieved .Finally, the rearranged images are returned to the user , a method of improving the resource allocation matrix is proposed as well .Experimental results show that the algorithm not only has a strong robustness but also has an excellent feedback effect .