非极大值抑制算法（NMS）

最新推荐文章于 2024-06-27 18:03:04 发布

sinat_29552923

最新推荐文章于 2024-06-27 18:03:04 发布

阅读量6.4k

点赞数 1

分类专栏：机器学习文章标签：算法机器学习

机器学习专栏收录该内容

6 篇文章 3 订阅

订阅专栏

1. 算法原理

非极大值抑制算法（Non-maximum suppression, NMS）的本质是搜索局部极大值，抑制非极大值元素。

2. 3邻域情况下NMS的实现

3邻域情况下的NMS即判断一维数组I[W]的元素I[i](2<=i<=W-1)是否大于其左邻元素I[i-1]和右邻元素I[i+1]，算法流程如下图所示：

a. 算法流程3-5行判断当前元素是否大于其左邻与右邻元素，如符合条件，该元素即为极大值点。对于极大值点I[i]，已知I[i]>I[i+1]，故无需对i+1位置元素做进一步处理，直接跳至i+2位置，对应算法流程第12行。

b. 若元素I[i]不满足算法流程第3行判断条件，将其右邻I[i+1]作为极大值候选，对应算法流程第7行。采用单调递增的方式向右查找，直至找到满足I[i]>I[i+1]的元素，若i<=W-1，该点即为极大值点，对应算法流程第10-11行。

3. NMS在物体检测中的应用

物体检测中应用NMS算法的主要目的是消除多余（交叉重复）的窗口，找到最佳物体检测位置。

如上图所示，人脸检测中，虽然每个窗口均检测到人脸，但仅需给出一个最有可能表征人脸的窗口。

4. 算法程序

 
          function 
          pickLocate = nms(boxes, overlap) 
         
          % Non-maximum suppression. 
         
          % In object detect algorithm, select high score detections and skip windows 
         
          % covered by a previously selected detection. 
         
          % 
         
          % input - boxes : object detect windows. 
         
          %                 xMin yMin xMax yMax score. 
         
          %         overlap : suppression threshold. 
         
          % output - pickLocate : number of local maximum score. 
         
          boxes =  
          double 
          (boxes); 
         
          if 
          isempty 
          (boxes) 
         
          pickLocate = []; 
         
          else 
         
          xMin = boxes(:, 1); 
         
          yMin = boxes(:, 2); 
         
          xMax = boxes(:, 3); 
         
          yMax = boxes(:, 4); 
         
          s = boxes(:,  
          end 
          ); 
         
          % area of every detected windows. 
         
          area 
          = (xMax - xMin + 1) .* (yMax - yMin + 1); 
         
          % sort detected windows based on the score. 
         
          [vals, I] =  
          sort 
          (s); 
         
          pickLocate = []; 
         
          while 
          ~ 
          isempty 
          (I) 
         
          last 
          =  
          length 
          (I); 
         
          i 
          = I( 
          last 
          ); 
         
          pickLocate = [pickLocate;  
          i 
          ]; 
         
          suppress = [ 
          last 
          ]; 
         
          for 
          pos = 1 :  
          last 
          - 1 
         
          j 
          = I(pos);   
         
          % covered area. 
         
          xx1 =  
          max 
          (xMin( 
          i 
          ), xMin( 
          j 
          )); 
         
          yy1 =  
          max 
          (yMin( 
          i 
          ), yMin( 
          j 
          )); 
         
          xx2 =  
          min 
          (xMax( 
          i 
          ), xMax( 
          j 
          )); 
         
          yy2 =  
          min 
          (yMax( 
          i 
          ), yMax( 
          j 
          )); 
         
          w = xx2 - xx1 + 1; 
         
          h = yy2 - yy1 + 1; 
         
          if 
          ((w > 0) && (h > 0)) 
         
          % compute overlap. 
         
          o = w * h /  
          min 
          ( 
          area 
          ( 
          i 
          ),  
          area 
          ( 
          j 
          )); 
         
          if 
          (o > overlap) 
         
          suppress = [suppress; pos]; 
         
          end 
         
          end 
         
          % xx1 = max(x1(i), x1(I(1:last-1))); 
         
          % yy1 = max(y1(i), y1(I(1:last-1))); 
         
          % xx2 = min(x2(i), x2(I(1:last-1))); 
         
          % yy2 = min(y2(i), y2(I(1:last-1))); 
         
          % w = max(0.0, xx2-xx1+1); 
         
          % h = max(0.0, yy2-yy1+1); 
         
          % inter = w.*h; 
         
          % o = inter ./ (area(i) + area(I(1:last-1)) - inter); 
         
          % saving the windows which o less than threshold. 
         
          % I = I(o <= overlap); 
         
          end 
         
          I(suppress) = []; 
         
          end 
         
          end