目的 为了能够得到图像更加丰富的纹理特征,提出一种新的自适应加权局部二值模式算法。方法 首先,将图像进行分块,利用新算法提取每个子块的局部二值模式的纹理直方图;然后,将各子图像的信息熵作为直方图的加权依据,对每个子块对应的直方图进行自适应加权,并将所有子块的直方图连接成最终的纹理特征。提取每个子块的局部纹理特征时的方法为:以某一像素点为中心取相邻的8个像素组成一个局部邻域,在该邻域内依据自适应设定的阈值分别比较3对水平方向和3对竖直方向像素值的大小,以此获得6位二进制码并将每位二进制码乘以相应的权重后相加,累加和即为该邻域新的局部二值模式纹理特征。结果 在两大人脸数据库上进行的实验结果表明,利用本文提出的方法提取纹理特征,并结合最近邻分类法可以得到85.29%和96.50%的正确识别率。结论 文中提出的自适应加权局部二值模式特征能够获取图像中更加丰富的纹理信息,因而具有较高的正确识别率,并且对于其他的物体识别也具有一定的参考价值。

Objective To achieve an abundant texture, a new weighted local binary pattern (W-LBP) is proposed in this study. Method An image is divided into several sub-images. W-LBP texture histograms are extracted from each sub-image. The proposed algorithm adaptively weights the W-LBP histograms of sub-patches on the basis of their information entropy and serially connects all histograms to create a final texture feature. The process of extracting the W-LBP texture of each sub-image is as follows. First, a local neighborhood is constructed with a certain centered pixel and surrounding eight pixels. A string of binary codes with six bits is then obtained by comparison of three pairs of horizontal and vertical pixels within the local neighborhood with an adaptive threshold. The texture feature of the local region is the sum obtained from the addition of each bit weighted by some weighting factor. Therefore, after the entire sub-image is traversed, the corresponding feature value of each pixel can be calculated. All the values can generate a statistical histogram, which is regarded as the feature value of the sub-image. Result Experimental results from the two famous face databases indicate that the proposed method with the nearest neighbor classification obtains correct recognition rates of 85.29% and 96.50%. Conclusion In terms of obtaining abundant texture information, the proposed feature in this study can lead to a high face recognition rate. The new feature also has some reference values for object recognition in other fields.