python 新的随机爬山算法，一起交流~

前言

最近在做聚焦算法，需要找到无参考图像中图像质量最好的，其中用到了随机爬山算法，
最开始采用三轮爬山，三轮步长不同，寻找最优解，但是发现精度不够，有的时候因为步长原因会跳过最佳点
后采用新的随机爬山算法

原理介绍

本算法第一轮采用大步长爬山算法确定范围值，如果找到局部峰值后，再向前继续找一步，如果同为下降趋势则停止
后几轮采用小步长迭代方法寻找最优解，最优解结果范围：±最后一轮步长，如果对迭代次数没有要求，最后步长为1迭代会更精确
流程图如下所示

爬山算法示意图

如图根据三轮后选择峰值点。

核心代码

class AutoFEL():

	def run_procedure(self,img,cur_fel):
	
        if  self.curCircle == 1: # 1轮采用爬山
            self.x_list.append(int(cur_fel))
            var = self.img_var(img)
            focus_v = self.focus_value(img, 3, 64, 5, 5)
            scharr_v = self.scharr_demo(img)
            result_v = np.average((var, focus_v, scharr_v), weights=[4, 3, 3])
            # print('value = ',result_v)
            self.result_list.append(result_v)
					
			#先处理前两张图像分辨前进方向 
            if self.imageCount < 2:
                if self.imageCount == 0:
                    if cur_fel + self.stepsize <= 200:
                        nextFel = cur_fel + self.stepsize * self.direction
                        self.ret_status = 0
                        self.imageCount += 1
                        return nextFel, self.ret_status
                    else:
                        self.direction = -1
                        nextFel = cur_fel + self.stepsize * self.direction
                        self.ret_status = 0
                        self.imageCount += 1
                        return nextFel, self.ret_status
                else:
                    # check the direction
                    if self.result_list[0] >= self.result_list[1] and cur_fel - self.stepsize * 2 > -200:
                        self.direction = -1
                        self.result_list.pop(0)
                        self.x_list.pop(0)

                    nextFel = cur_fel + self.stepsize * self.direction
                    self.ret_status = 0
                    self.imageCount += 1
                    return nextFel, self.ret_status
            else:

                if self.isToPeak :

                    if self.result_list[-1] < self.result_list[-2]:
                        # 连续下降，找到峰值，以及当前爬坡结束，
                        nextFel = self.peakFel - 2 * self.stepsize
                        print("nextFel = ",nextFel)
                        self.small_range = self.peakFel + self.stepsize * 2
                        print("small_range = ",self.small_range)
                        self.stepsize = int(self.stepsize / 2)
                        self.x_list = []
                        self.result_list = []
                        self.curCircle += 1
                        self.ret_status = 0
                        return nextFel, self.ret_status
                    else:
                        # 非连续下降，isToPeak = False， 继续沿着原来的方向走
                        nextFel = cur_fel + self.stepsize * self.direction
                        self.ret_status = 0
                        self.isToPeak = False
                        return nextFel, self.ret_status

                else:
                    state_flag, max_x = self.jugde_max(self.x_list, self.result_list)
                    if state_flag == 1:
                        self.peakFel = max_x
                        print("p = ",self.peakFel)
                        nextFel = cur_fel + self.stepsize * self.direction
                        self.isToPeak = True
                        self.ret_status = 0
                        return nextFel, self.ret_status
                    else:
                        nextFel = cur_fel + self.stepsize * self.direction
                        self.ret_status = 0
                        return nextFel, self.ret_status
                        
        else: # 2, 3轮根据固定步长遍历

            if cur_fel < self.small_range:
                self.x_list.append(int(cur_fel))
                var = self.img_var(img)
                focus_v = self.focus_value(img, 3, 64, 5, 5)
                scharr_v = self.scharr_demo(img)
                result_v = np.average((var, focus_v, scharr_v), weights=[4, 3, 3])
                # print('value = ',result_v)
                self.result_list.append(result_v)
                nextFel = cur_fel + self.stepsize
                self.ret_status = 0
                return nextFel, self.ret_status
            else:

                max_value = max(self.result_list)
                max_t = self.result_list.index(max_value)
                max_x = self.x_list[max_t]
                self.peakFel = max_x
                if self.curCircle != 3:
                    nextFel = self.peakFel - 2 * self.stepsize

                    self.x_list = []
                    self.result_list = []
                    self.curCircle += 1
                    self.ret_status = 0
                    self.small_range = self.peakFel + 2 * self.stepsize
                    self.stepsize = 4
                    return nextFel, self.ret_status
                else:
                    self.ret_status = 1
                    return self.peakFel,self.ret_status