iOS图像畸变校正：如何实现类似的线条拉直效果？

核心问题分析 #

你使用CIPerspectiveCorrection、CIPerspectiveTransform滤镜未达到预期效果，核心原因通常是透视变换的顶点坐标不精准，或是未正确处理Core Image与UIKit的坐标系差异。

解决方案步骤 #

1. 精准获取校正区域顶点坐标 #

透视类滤镜的效果完全依赖四个源顶点（对应待校正区域的四个边角）的准确性，手动选点误差大，建议用Vision框架自动检测：

• 创建矩形检测请求：

  let rectangleRequest = VNDetectRectanglesRequest { request, error in
      guard let results = request.results as? [VNRectangleObservation], let rectangle = results.first else { return }
      // 检测到的矩形顶点（Vision坐标系：图像像素原点在左下角）
      let topLeft = rectangle.topLeft
      let topRight = rectangle.topRight
      let bottomLeft = rectangle.bottomLeft
      let bottomRight = rectangle.bottomRight
      // 直接传入校正方法
      self.applyPerspectiveCorrection(with: [topLeft, topRight, bottomLeft, bottomRight], image: originalImage)
  }
  // 设置检测参数，匹配图像内容的宽高比范围
  rectangleRequest.minimumAspectRatio = 1.0
  rectangleRequest.maximumAspectRatio = 2.0
  

• 执行检测：

  guard let ciImage = CIImage(image: originalImage) else { return }
  let handler = VNImageRequestHandler(ciImage: ciImage)
  try? handler.perform([rectangleRequest])
  

2. 正确应用CIPerspectiveTransform滤镜 #

若手动选点，需将UIKit左上角原点的坐标转换为Core Image左下角原点的坐标：

func convertUIPointToCIPoint(_ point: CGPoint, imageSize: CGSize) -> CGPoint {
    return CGPoint(x: point.x, y: imageSize.height - point.y)
}

func applyPerspectiveCorrection(with vertices: [CGPoint], image: UIImage) {
    guard let ciImage = CIImage(image: image) else { return }
    let filter = CIFilter(name: "CIPerspectiveTransform")!
    
    filter.setValue(CIVector(cgPoint: vertices[0]), forKey: "inputTopLeft")
    filter.setValue(CIVector(cgPoint: vertices[1]), forKey: "inputTopRight")
    filter.setValue(CIVector(cgPoint: vertices[2]), forKey: "inputBottomLeft")
    filter.setValue(CIVector(cgPoint: vertices[3]), forKey: "inputBottomRight")
    filter.setValue(ciImage, forKey: kCIInputImageKey)
    
    guard let outputImage = filter.outputImage else { return }
    let context = CIContext(options: [.useSoftwareRenderer: false])
    if let cgImage = context.createCGImage(outputImage, from: outputImage.extent) {
        let correctedImage = UIImage(cgImage: cgImage)
        // 使用校正后的图像
    }
}

3. 替代方案：CIPerspectiveCorrection自动校正 #

该滤镜需指定目标矩形（校正后的区域），若自动校正失效，需确保目标矩形与源区域宽高比匹配：

func applyAutoPerspectiveCorrection(image: UIImage, sourceVertices: [CGPoint]) {
    guard let ciImage = CIImage(image: image) else { return }
    let filter = CIFilter(name: "CIPerspectiveCorrection")!
    
    // 目标矩形设为图像完整尺寸
    let targetRect = CIVector(cgRect: ciImage.extent)
    filter.setValue(targetRect, forKey: "inputTargetRect")
    // 传入检测到的四个源顶点
    filter.setValue(CIVector(cgPoint: sourceVertices[0]), forKey: "inputTopLeft")
    filter.setValue(CIVector(cgPoint: sourceVertices[1]), forKey: "inputTopRight")
    filter.setValue(CIVector(cgPoint: sourceVertices[2]), forKey: "inputBottomLeft")
    filter.setValue(CIVector(cgPoint: sourceVertices[3]), forKey: "inputBottomRight")
    filter.setValue(ciImage, forKey: kCIInputImageKey)
    
    // 后续处理同CIPerspectiveTransform
}

常见排查点 #

• 坐标系错误：务必确认顶点使用Core Image/Vision的左下角原点坐标系，而非UIKit的左上角。
• 顶点偏差：四个源顶点必须严格对应待校正区域的真实边角，微小偏差都会导致校正效果不佳。
• 图像缩放问题：处理UIImage时需基于其像素尺寸（image.size * image.scale），而非点尺寸。

内容的提问来源于stack exchange，提问作者xKore Evolution