一种自适应特征分块加权的图像检索方法

文中提出了一种自适应特征加权的图像检索方法。首先提取图像的颜色直方图以及梯度直方图，根据图像的特点，用不同的方法进行分割。然后将已分割图像分成4×4小块，按照其颜色和纹理的分布情况，计算每小块的特征加权值。最后，以分块加权的一维颜色矢量直方图为特征，进行相似度匹配，欧氏距离较小的图像为检索结果。实验结果表明，综合利用颜色和边缘特征比单独用一个特征进行图像检索的效果要好，自适应特征加权后的检索准确率更高，对于拥有较明显的目标与背景差异的图像，该方法特别有效。

A Block-based Methods for Image Retrieval Using Adaptive Feature Weight

An image retrieval method based on adaptive feature weight is proposed in this paper. First of all ,extract the image color histogram and gradient histogram, according to the characteristics of the image, segmentation is conducted in a different way. Then the image segmented is divided into small pieces of 4 × 4 ,according to the distribution of its color and texture, calculate every small piece of the weighted value. Finally, one-dimensional weighted color histogram vector as feature, carry on similarity matching, the image with smaller Euclidean distance as the results. Experiments show that comprehensive utilization of color and edge features is better than with one feature for image retrieval, the retrieval accuracy after the adaptive feature weighting is higher, especially for those image with distinct target and background,this method is effective.