基于哈希算法的图像拷贝检测

在局部保持投影算法（LPP）及最大熵模型的基础上提出了一种基于哈希算法的图像拷贝检测算法。该算法通过对LPP进行优化,可以对已有的图像特征进行快速降维处理,并避免过拟合现象的出现。同时,将低维度特征映射为紧凑的二进制哈希码,大大减少了存储和计算开销,能在千万级图像数据库上实现高效拷贝检测。通过实验确定了使拷贝检测质量最优的哈希码长度和汉明码。实验结果表明,该算法能够获得很好的拷贝检测效果。

Hashing-based Image Copy Detection

On the basis of Locality Preserving Projection（LPP） and Maximum Entropy, a hashing-based algorithm for image copy detection is proposed. An optimized version of LPP is used to map existing high-dimensional features of images into low-dimensional subspace without over-fitting. Next, the low-dimensional features are mapped into compact binary hash codes, which greatly reduces the cost of storage and computation, making it possible to accomplish efficient copy detection on database with more million images. The optimal length of hash codes and hamming distance through experiment are chosen. Experimental results indicate that the proposed approach can bring excellent performance of image copy detection.