中文摘要：

论文针对视觉词袋（BOVW）模型放弃图像空间结构的缺点,提出一种基于Hesse稀疏编码的图像检索算法。首先,建立n-words模型,获得图像局部特征表示。n-words模型由一系列连续视觉词获得,是图像特征的一种高级描述。该文从n=1到n=5进行试验,寻找最恰当的n值;其次,将二阶Hesse能量函数融入标准稀疏编码的目标函数,得到Hesse稀疏编码公式;最后,以获得的n-words序列作为编码特征,利用特征符号搜索算法求解最优Hesse系数,计算相似度,返回检索结果。实验在两类数据集上进行,与BOVW模型和已有的算法相比,新算法极大地提高了图像检索的准确率。

英文摘要：

To deal with the problem that the Bag-Of-Visual-Words（BOVW） model discards image spatial structure, a new method based on the Hessian sparse coding for image retrieval is introduced. First, the n-words model is built in order to obtain the local feature representation. The n-words model can establish a high-level description using a series of visual word sequences to represent an image. The experiments are performed from n=1 to n=5 to seek the proper n. Second, the Hessian sparse coding formulation is acquired by incorporating the Hessian energy function into the standard sparse coding formulation. Finally, using the obtained n-words sequences as the encoding features, the optimal Hessian coefficients are calculated through the feature-sign search algorithm. The similarity is computed and the retrieval results are returned. The experiments are performed on the two datasets, the results show that the proposed new method for image retrieval outperforms the BOVW model and existent methods.