基于多决策函数的复杂背景下自动皮肤检测

Automatic Skin Detection in Complex Background Using Multiple Decision Function

皮肤检测被广泛用于计算机视觉各种应用场景。然而,皮肤检测受非线性照明、相机特性、成像条件和人种特征影响,使得高效、准确地检测皮肤仍然是个挑战。为优化检测效果,提出一种新颖的使用线性与非线性决策函数的皮肤检测算法。该方法先对皮肤进行颜色空间变换,提取皮肤像素值,并构建像素三维分布图。为了得到皮肤检测的线性与非线性决策函数,根据像素在三维分布图中的特点,将其投影到二维侧面,利用皮肤簇在二维坐标中的形状特性建立决策模型;最后使用准确率和精确率对检测模型性能进行数值化评估。在光照不均匀、背景区域与肤色相近以及皮肤有病斑条件下,皮肤检测平均准确率91.82%,平均精确率90.36%,表明提出的算法得到令人满意的效果,鲁棒性好,有望应用在皮肤病斑检测中。

Skin detection is widely used in various applications of computer vision.However,skin detection is affected by non-linear lighting,camera characteristics,imaging conditions and ethnic characteristics,so it is still a challenge to detect skin efficiently and accurately.To optimize the detection results,a novel skin detection algorithm using linear and non-linear decision functions is proposed in this paper.This method transforms the skin color space,extracts the skin pixel values,and then builds a three-dimensional distribution of the pixels.In order to get the linear and non-linear decision functions of skin detection,according to the characteristics of pixels in the three-dimensional distribution map,it is projected to the two-dimensional side,and the decision model is established by using the shape characteristics of skin clusters in two-dimensional coordinates.Finally,the accuracy and precision are used to evaluate the performance of the detection model.Under uneven lighting,similar background color and skin lesions,the average accuracy of skin detection is 91.82%,and the average precision of skin detection is 90.36%,which shows that the proposed algorithm has satisfactory results and good robustness,and highlights its potential application in skin lesion detection.