CAJ | 학술논문

图像基学习是图像特征提取与表示的重要方法之一.非负稀疏编码不仅具有标准稀疏编码算法的自适应性、空间的局部性、方向性和频域的带通性,而且更能反应哺乳动物的视觉机制.本文在非负稀疏编码的基础上,利用经验模态分解技术加入了图像的结构信息,提出了结合经验模态分解的非负稀疏编码算法,保证了系数矩阵的稀疏性与所提取图像特征的结构性.学习得到的图像基不仅具有非负稀疏编码的特征,而且更好地表示出图像的结构信息.
도상기학습시도상특정제취여표시적중요방법지일.비부희소편마불부구유표준희소편마산법적자괄응성、공간적국부성、방향성화빈역적대통성,이차경능반응포유동물적시각궤제.본문재비부희소편마적기출상,이용경험모태분해기술가입료도상적결구신식,제출료결합경험모태분해적비부희소편마산법,보증료계수구진적희소성여소제취도상특정적결구성.학습득도적도상기불부구유비부희소편마적특정,이차경호지표시출도상적결구신식.
Image base learning is one of the important ways of image feature extraction and image expression. Non-nega-tive sparse coding not only features the adaptability of standard sparse coding, spatial localization, orientation, and band-pass in different spatial frequency bands, but also responds to mammal's visual mechanism well. On the basis of the non-negative sparse coding, this paper joins the image structure information using the experience rnodality decomposition tech-nology, proposes a combination of EMD and the non-negative sparse coding algorithm, and ensures the sparseness of the co-efficient matrix and the structural characteristics of the image bases extracted. The learned image bases not only have the characteristic of non-negative sparse coding, but express the image's structure information well.