基于重心坐标乘积量化的图像检索方法

Barycentric coordinates product quantization for image retrieval

为减少在大范围图像检索过程中因硬编码导致的量化误差,提出一种基于重心坐标的软编码方法,利用乘积量化将高维特征空间分解成低维特征子空间的笛卡尔积的形式,分别进行量化,以重心坐标的形式进行稀疏表示,最小化产生的量化误差,使结果表示更接近于实际中的原始数据。通过在3个公开可得的图像数据集上的实验,验证了提出方法可有效提高ANN查询的精度。

To reduce the quantization error caused by hard coding in large scale image retrieval,an encoding scheme called barycentric coordinates based soft coding was put forward.Product quantization was used to decompose the high-dimensional feature space into a Cartesian product of low dimensional subspaces and each of them was quantized separately,minimizing the quantization errors,and making the resulting representation essentially close to the original data in practice.Experiments on three publicly available image datasets demonstrate that the proposed method improves the precision of query for ANN search effectively.