基于圖正則化與非負(fù)組稀疏的自動(dòng)圖像標(biāo)注
doi: 10.11999/JEIT141282
基金項(xiàng)目:
國(guó)家自然科學(xué)基金(61271439)資助課題
Automatic Image Annotation via Graph Regularization and Non-negative Group Sparsity
-
摘要: 設(shè)計(jì)一個(gè)穩(wěn)健的自動(dòng)圖像標(biāo)注系統(tǒng)的重要環(huán)節(jié)是提取能夠有效描述圖像語(yǔ)義的視覺特征。由于顏色、紋理和形狀等異構(gòu)視覺特征在表示特定圖像語(yǔ)義時(shí)所起作用的重要程度不同且同一類特征之間具有一定的相關(guān)性，該文提出了一種圖正則化約束下的非負(fù)組稀疏(Graph Regularized Non-negative Group Sparsity, GRNGS)模型來實(shí)現(xiàn)圖像標(biāo)注，并通過一種非負(fù)矩陣分解方法來計(jì)算其模型參數(shù)。該模型結(jié)合了圖正則化與l2,1-范數(shù)約束，使得標(biāo)注過程中所選的組群特征能體現(xiàn)一定的視覺相似性和語(yǔ)義相關(guān)性。在Corel5K和ESP Game等圖像數(shù)據(jù)集上的實(shí)驗(yàn)結(jié)果表明：相較于一些最新的圖像標(biāo)注模型，GRNGS模型的魯棒性更強(qiáng)，標(biāo)注結(jié)果更精確。
-
關(guān)鍵詞:
- 圖像標(biāo)注 /
- 圖正則化 /
- 組稀疏 /
- 非負(fù)矩陣分解
Abstract: Extracting an effective visual feature to uncover semantic information is an important work for designing a robust automatic image annotation system. Since different kinds of heterogeneous features (such as color, texture and shape) show different intrinsic discriminative power and the same kind of features are usually correlated for image understanding, a Graph Regularized Non-negative Group Sparsity (GRNGS) model for image annotation is proposed, which can be effectively solved by a new method of non-negative matrix factorization. This model combines graph regularization withl2,1-norm regularization, and is able to select proper group features, which can describe both visual similarities and semantic correlations when performing the task of image annotation. Experimental results reported over the Corel5K and ESP Game databases show the robust capability and good performance of the proposed method. -
計(jì)量
- 文章訪問數(shù): 2327
- HTML全文瀏覽量: 99
- PDF下載量: 1032
- 被引次數(shù): 0
下載:
