Abstract: For the problem of accuracy, realtime performance and illumination in image retrieval and feature extraction with SIFT, a new image features extracting algorithm based on the SIFT was proposed and the Local Binary Patterns (LBP) algorithm. It used the same keypoint detection method as SIFT. After getting the keypoints of the image features, the SIFTLBP descriptor was made up of statistics of gradient information in 16×16 region and rotation invariant LBP value in 9×9 regions around each keypoint, and then building SIFTLBP Feature Descriptor of image in each pixel region as the center. Finally, the extracted features of data were selected based on genetic algorithm, and removed redundant information of feature point to reduce the dimension of the feature vector. Experimental results showed that the proposed algorithm had a good matching result on Visual Image Feature Extraction, it was validated that the algorithm was strongly robust to changes in lighting conditions, and improved the accuracy and the speed of retrieval.

Key words: Scaleinvariant feature transform；Rotation invariant local binary patterns；Image retrieval；Feature extraction, Feature points

摘要： 针对尺度不变特征变换( S I F T) 算法在图像特征提取和检索中精度、 实时性以及对光照条件变化描述较差的问题, 提出了S I F T和局部二值模式( L B P) 相结合的图像特征提取算法。采用旋转不变L B P算法统计关键点周围1 6×1 6区域的梯度信息并计算周围9×9区域的L B P值, 以区域中每个像素点为中心构建图像的S I F T - L B P特征描述子。采用了基于遗传算法的特征选择方法, 剔除了特征点的冗余信息, 降低了特征向量维数。实验结果表明, S I F T - L B P算法具有良好的特征匹配效果, 对光照条件的变化具有较强的鲁棒性, 进一步提高了检索准确率和检索速度。

关键词: 尺度不变特征变换; , 旋转不变局部二值模式; , 图像检索; , 特征提取; , 特征点