在图像中找到边缘后可变调整线条粗细的好方法是什么？

Question

CIEdge（以及其他 CIFilters）在设备和 iOS 版本之间不再一致地工作。所以我决定用卷积滤波器寻找边缘。成功：我可以得到一个很好的图像黑白蒙版，让渐变或纯色显示出来：

问题是，我想不出一个可变调整线条粗细的好方法。我尝试的一种方法是降低源图像的一个版本，并在黑白相机图像之上应用卷积和放大。那种作品，但线条看起来锯齿状和粗糙。有任何想法吗？这是我的代码：

func sketch(with ciImage: CIImage) -> CIImage {

    var sourceCore = ciImage

    var convolutionValue_A:CGFloat = -0.0925937220454216
    var convolutionValue_B:CGFloat = -0.4166666567325592
    var convolutionValue_C:CGFloat = -1.8518532514572144
    var convolutionValue_D:CGFloat = 0.23148006200790405
    var convolutionValue_E:CGFloat = 4.5833334922790527
    var convolutionValue_F:CGFloat = 14.166666984558105

    var brightnessVal:CGFloat = 1.1041666269302368
    var contrastVal:CGFloat = 3.0555555820465088

    // radially symmetrical convolution weights:
    var weightsArr: [CGFloat] = [
        convolutionValue_A, convolutionValue_A, convolutionValue_B, convolutionValue_B, convolutionValue_B, convolutionValue_A, convolutionValue_A,
        convolutionValue_A, convolutionValue_B, convolutionValue_C, convolutionValue_C, convolutionValue_C, convolutionValue_B, convolutionValue_A,
        convolutionValue_B, convolutionValue_C, convolutionValue_D, convolutionValue_E, convolutionValue_D, convolutionValue_C, convolutionValue_B,
        convolutionValue_B, convolutionValue_C, convolutionValue_E, convolutionValue_F, convolutionValue_E, convolutionValue_C, convolutionValue_B,
        convolutionValue_B, convolutionValue_C, convolutionValue_D, convolutionValue_E, convolutionValue_D, convolutionValue_C, convolutionValue_B,
        convolutionValue_A, convolutionValue_B, convolutionValue_C, convolutionValue_C, convolutionValue_C, convolutionValue_B, convolutionValue_A,
        convolutionValue_A, convolutionValue_A, convolutionValue_B, convolutionValue_B, convolutionValue_B, convolutionValue_A, convolutionValue_A
    ]

    let inputWeights:CIVector = CIVector(values: weightsArr, count: weightsArr.count)

    sourceCore = sourceCore
        .applyingFilter("CIColorControls", parameters: [kCIInputImageKey: sourceCore,
                                                        kCIInputSaturationKey: 0.0,
                                                        kCIInputBrightnessKey: brightnessVal,
                                                        kCIInputContrastKey: contrastVal])

    // transforms image to only show edges in black and white
    sourceCore = sourceCore
        .applyingFilter("CIConvolution7X7", parameters: [kCIInputImageKey: sourceCore,

    let whiteCIColor = CIColor.white
    let whiteColor = CIImage(color: whiteCIColor).cropped(to: ciImage.extent)

    // for some reason, I need to blend the black and white mask with the color white.
    // either CIColorDodgeBlendMode or CILinearDodgeBlendMode seems to work fine here:
    sourceCore = sourceCore
            .applyingFilter("CIColorDodgeBlendMode", parameters: [kCIInputImageKey: sourceCore,
                                                                    kCIInputBackgroundImageKey: whiteColor])
                                                         kCIInputWeightsKey: inputWeights]).cropped(to: sourceCore.extent)

    // give camera image a black and white Noir effect
    var ciImage = ciImage
        .applyingFilter("CIPhotoEffectNoir", parameters: [kCIInputImageKey: ciImage])


    // make solid color
    let color = CIColor(red: 0.819, green: 0.309, blue: 0.309)
    let colFilter = CIFilter(name: "CIConstantColorGenerator")!
    colFilter.setValue(color, forKey: kCIInputColorKey)
    var solidColor = colFilter.value(forKey: "outputImage") as! CIImage
    solidColor = solidColor.cropped(to: ciImage.extent)

    // color is shown through outlines correctly, 
    // and image is black and white
    sourceCore = sourceCore
        .applyingFilter("CIBlendWithMask", parameters: [
            kCIInputImageKey: ciImage, // black and white image
            kCIInputBackgroundImageKey: solidColor, // solid color
            kCIInputMaskImageKey:sourceCore]) // edge work image

    ciImage = sourceCore

    return ciImage

}

Answer 1

边缘的厚度取决于卷积核的宽度或图像分辨率。

要（动态）增加卷积的宽度，您可能需要实现自己的自定义 CIKernel，因为 Core Image 仅支持最多 9x9 的卷积核。更宽的内核应该提供更平滑的结果，但也更昂贵。

您可以通过在应用内核之前缩小图像（就像您已经做的那样）并在之后再次放大来实现类似的效果。您可以尝试使用 "smart" 放大过滤器，例如 CILanczosScaleTransform 或 CIBicubicScaleTransform 并使用它们的参数。使用 inputC 的双三次滤波器，您应该能够控制结果的柔和度。

在图像中找到边缘后可变调整线条粗细的好方法是什么？

What would be a good method to variably adjust line thickness after finding edges in an image?

cifilter

swift