CAJ | 학술논문

针对BOVW模型忽略图像特征空间排列导致量化误差较大的缺点,利用角点和特征点对图像进行区域分割,结合区域的空间排列信息,提出一种多通道融合的图像检索方法. 其主要思想是将子区域编码和特征空间排列直方图结合组建视觉短语,这种构造方式在减少编码误差的同时还能更好地保留局部空间信息. 首先,利用稀疏编码保留局部信息的高效性对提取的子区域进行编码;其次,利用特征的空间位置关系,计算子区域内的特征空间排列直方图;利用区域编码和特征排列直方图构建视觉短语;最后,结合BOVW模型的鲁棒性,统计视觉短语直方图用于图像检索. 实验结果表明,该检索方法不仅比BOVW和SPMBOVM有更好的检索准确率,而且其编码过程稳定,误差较小.
침대BOVW모형홀략도상특정공간배렬도치양화오차교대적결점,이용각점화특정점대도상진행구역분할,결합구역적공간배렬신식,제출일충다통도융합적도상검색방법. 기주요사상시장자구역편마화특정공간배렬직방도결합조건시각단어,저충구조방식재감소편마오차적동시환능경호지보류국부공간신식. 수선,이용희소편마보류국부신식적고효성대제취적자구역진행편마;기차,이용특정적공간위치관계,계산자구역내적특정공간배렬직방도;이용구역편마화특정배렬직방도구건시각단어;최후,결합BOVW모형적로봉성,통계시각단어직방도용우도상검색. 실험결과표명,해검색방법불부비BOVW화SPMBOVM유경호적검색준학솔,이차기편마과정은정,오차교소.
The BOVW model ignores the image feature spatial arrangement, thus causing quantization error. Con?sidering this shortcoming, we divided an image into a series of sub?regions according to corners and features. Com?bining spatial arrangement information of the sub?regions, we, using multimodal fusion, proposed a new image re?trieval method. The main idea is to construct visual phrases through sub?region encoding and feature spatial arrange?ment histograms. By this combination, it not only reduces the encoding error but also better preserves the local spa?tial information. First, using the advantages of sparse coding, we encoded the sub?regions; second, according to the feature spatial location relations, sub?region feature spatial arrangement histograms were calculated; third, visual phrases were composed of sub?region encoding and feature spatial arrangement histograms;at last, incorpora?ting the robustness of BOVW model, we calculated the visual phrase histograms for image retrieval. The results and their analysis show preliminarily that the proposed retrieval method not only has better retrieval accuracy than BOVW and SPMBOVW but also its encoding is more stable and the error is smaller.