TPS 薄板样条插值 python的opencv实现_cv2.createthinplatesplineshapetransformer()

原始论文：

Donato, G., & Belongie, S. J. (2003). Approximation methods for thin plate spline mappings and principal warps. Department of Computer Science and Engineering, University of California, San Diego.

效果：

原图：

变化后：

Code：

import cv2
import numpy as np
import random
# 首先读入img
img = cv2.imread('data/liv.png',cv2.IMREAD_COLOR)
img = cv2.resize(img,(180,32))
# N对基准控制点
N=5
points=[]
dx=int(180/(N-1))
for i in range(2*N):
    points.append((dx*i,4))
    points.append((dx*i,36))
# 周围拓宽一圈
img = cv2.copyMakeBorder(img,4,4,0,0,cv2.BORDER_REPLICATE)
# 画上绿色的圆圈
# for point in points:
# 	cv2.circle(img, point, 1, (0, 255, 0), 2)
tps = cv2.createThinPlateSplineShapeTransformer()
 
sourceshape = np.array(points,np.int32)
sourceshape=sourceshape.reshape(1,-1,2)
matches =[]
for i in range(1,N+1):
    matches.append(cv2.DMatch(i,i,0))
 
# 开始随机变动
newpoints=[]
PADDINGSIZ=10
for i in range(N):
    nx=points[i][0]+random.randint(0,PADDINGSIZ)-PADDINGSIZ/2
    ny=points[i][1]+random.randint(0,PADDINGSIZ)-PADDINGSIZ/2
    newpoints.append((nx,ny))
print(points,newpoints)
targetshape = np.array(newpoints,np.int32)
targetshape=targetshape.reshape(1,-1,2)
tps.estimateTransformation(sourceshape,targetshape ,matches)
img=tps.warpImage(img)
cv2.imwrite('tmp.png',img)
 

Debug参考：

https://xbuba.com/questions/41536344

https://docs.opencv.org/3.4/df/dfe/classcv_1_1ShapeTransformer.html

https://qiita.com/SousukeShimoyama/items/2bf8defb2d057bb8b742#tps%E3%81%AE%E3%82%A4%E3%83%B3%E3%82%B9%E3%82%BF%E3%83%B3%E3%82%B9%E3%82%92%E7%94%9F%E6%88%90