CAJ | 학술논문

在纹理分类应用背景下,原始韦伯局部描述符( WLD)对纹理模式区分能力有限。针对该问题,提出一种基于正负梯度改进的WLD( WLD-PNG)。利用局部窗内像素点间灰度变化的正负梯度构建纹理特征描述符,通过分离计算正负梯度的差分激励算子,保留灰度等级变化的正负性信息,以增强纹理模式的可区分性,运用均匀局部二值模式( uLBP)提取灰度等级变化的空间分布结构信息,并提高纹理模式的识别能力,使用均匀量化和编码技术将差分激励算子与uLBP结合,从而描述图像的纹理特征。在Brodatz和KTH-TIPS2-a纹理库上进行对比实验,结果表明,与原始WLD,uLBP,WLD+uLBP及已有改进的WLD等方法相比,WLD-PNG在提高纹理分类性能的同时,具有较好的稳健性和较低的计算复杂度。
재문리분류응용배경하,원시위백국부묘술부( WLD)대문리모식구분능력유한。침대해문제,제출일충기우정부제도개진적WLD( WLD-PNG)。이용국부창내상소점간회도변화적정부제도구건문리특정묘술부,통과분리계산정부제도적차분격려산자,보류회도등급변화적정부성신식,이증강문리모식적가구분성,운용균균국부이치모식( uLBP)제취회도등급변화적공간분포결구신식,병제고문리모식적식별능력,사용균균양화화편마기술장차분격려산자여uLBP결합,종이묘술도상적문리특정。재Brodatz화KTH-TIPS2-a문리고상진행대비실험,결과표명,여원시WLD,uLBP,WLD+uLBP급이유개진적WLD등방법상비,WLD-PNG재제고문리분류성능적동시,구유교호적은건성화교저적계산복잡도。
Aiming at the shortage of the discriminative ability to texture patterns for image texture classification using Weber Local Descriptor(WLD),an improved WLD based on Positive and Negative Gradient Features(WLD-PNG) is proposed. Positive and negative gradients are characterized by computing positive and negative differential excitations for preserving signed grayscale change information,and local texture structure information is represented by uniform Local Binary Patterns(uLBP). Combine both of them to build the image texture feature. The comparing experiments on the Brodatz and KTH-TIPS2-a texture databases demonstrate that WLD-PNG improves the distinctiveness of texture patterns, and has better robustness and lower computational complexity compared with original WLD, uLBP, WLD+uLBP and other improved WLD methods,etc.