Metal每日分享，圖像閥值素描濾鏡效果

本案例的目的是理解如何用Metal實現圖像閥值素描濾鏡，用於圖像閥值素描，形成有噪點的素描；

---

效果圖

Demo

• HarbethDemo地址

實操代碼

// 用於圖像閥值素描，形成有噪點的素描
let filter = C7ThresholdSketch.init(edgeStrength: 2.5, threshold: 0.25)

// 方案1:
let dest = BoxxIO.init(element: originImage, filter: filter)
ImageView.image = try? dest.output()

dest.filters.forEach {
    NSLog("%@", "\($0.parameterDescription)")
}

// 方案2:
ImageView.image = try? originImage.make(filter: filter)

// 方案3:
ImageView.image = originImage ->> filter

實現原理

• 過濾器

這款濾鏡採用並行計算編碼器設計.compute(kernel: "C7ThresholdSketch")，參數因子[edgeStrength, threshold]

對外開放參數

• threshold: 任何高於這個閾值的邊緣都是黑色的，低於白色的任何東西都是黑色的。
• edgeStrength: 調整濾波器的動態範圍。更高的值會導致更強的邊緣，但可以飽和強度的色彩空間。

public struct C7ThresholdSketch: C7FilterProtocol {
    
    public static let range: ParameterRange<Float, Self> = .init(min: 0.0, max: 1.0, value: 0.25)
    
    /// Any edge above this threshold will be black, and anything below white. Ranges from 0.0 to 1.0
    @ZeroOneRange public var threshold: Float = range.value

    /// Adjusts the dynamic range of the filter.
    /// Higher values lead to stronger edges, but can saturate the intensity colorspace.
    public var edgeStrength: Float = 1
    
    public var modifier: Modifier {
        return .compute(kernel: "C7ThresholdSketch")
    }
    
    public var factors: [Float] {
        return [edgeStrength, threshold]
    }
    
    public init(edgeStrength: Float = 1, threshold: Float = range.value) {
        self.edgeStrength = edgeStrength
        self.threshold = threshold
    }
}

• 着色器

取出周邊1像素對應點的歸一化座標，獲取到這些點對應的紅色值；
水平方向取頂部和底部紅色值做個整合處理，豎直方向取左邊和右邊做個整合處理，分別得到(h, v);
length計算出範圍值，step計算出閾值對應的顏色，最後得到黑色或白色像素顏色即可；

kernel void C7ThresholdSketch(texture2d<half, access::write> outputTexture [[texture(0)]],
                              texture2d<half, access::sample> inputTexture [[texture(1)]],
                              constant float *edgeStrength [[buffer(0)]],
                              constant float *threshold [[buffer(1)]],
                              uint2 grid [[thread_position_in_grid]]) {
    constexpr sampler quadSampler(mag_filter::linear, min_filter::linear);
    
    const float x = float(grid.x);
    const float y = float(grid.y);
    const float width = float(inputTexture.get_width());
    const float height = float(inputTexture.get_height());
    
    const float2 leftCoordinate = float2((x - 1) / width, y / height);
    const float2 rightCoordinate = float2((x + 1) / width, y / height);
    const float2 topCoordinate = float2(x / width, (y - 1) / height);
    const float2 bottomCoordinate = float2(x / width, (y + 1) / height);
    const float2 topLeftCoordinate = float2((x - 1) / width, (y - 1) / height);
    const float2 topRightCoordinate = float2((x + 1) / width, (y - 1) / height);
    const float2 bottomLeftCoordinate = float2((x - 1) / width, (y + 1) / height);
    const float2 bottomRightCoordinate = float2((x + 1) / width, (y + 1) / height);
    
    const half leftIntensity = inputTexture.sample(quadSampler, leftCoordinate).r;
    const half rightIntensity = inputTexture.sample(quadSampler, rightCoordinate).r;
    const half topIntensity = inputTexture.sample(quadSampler, topCoordinate).r;
    const half bottomIntensity = inputTexture.sample(quadSampler, bottomCoordinate).r;
    const half topLeftIntensity = inputTexture.sample(quadSampler, topLeftCoordinate).r;
    const half topRightIntensity = inputTexture.sample(quadSampler, topRightCoordinate).r;
    const half bottomLeftIntensity = inputTexture.sample(quadSampler, bottomLeftCoordinate).r;
    const half bottomRightIntensity = inputTexture.sample(quadSampler, bottomRightCoordinate).r;
    
    half h = -topLeftIntensity - 2.0h * topIntensity - topRightIntensity + bottomLeftIntensity + 2.0h * bottomIntensity + bottomRightIntensity;
    h = max(0.0h, h);
    half v = -bottomLeftIntensity - 2.0h * leftIntensity - topLeftIntensity + bottomRightIntensity + 2.0h * rightIntensity + topRightIntensity;
    v = max(0.0h, v);
    
    half mag = length(half2(h, v)) * half(*edgeStrength);
    mag = 1.0h - step(half(*threshold), mag);
    
    const half4 outColor = half4(half3(mag), 1.0h);
    outputTexture.write(outColor, grid);
}

Harbeth功能清單

• 支持ios系統和macOS系統
• 支持運算符函數式操作
• 支持多種模式數據源 UIImage, CIImage, CGImage, CMSampleBuffer, CVPixelBuffer.
• 支持快速設計濾鏡
• 支持合併多種濾鏡效果
• 支持輸出源的快速擴展
• 支持相機採集特效
• 支持視頻添加濾鏡特效
• 支持矩陣卷積
• 支持使用系統 MetalPerformanceShaders.
• 支持兼容 CoreImage.
• 濾鏡部分大致分爲以下幾個模塊：
  • Blend：圖像融合技術
  • Blur：模糊效果
  • Pixel：圖像的基本像素顏色處理
  • Effect：效果處理
  • Lookup：查找表過濾器
  • Matrix: 矩陣卷積濾波器
  • Shape：圖像形狀大小相關
  • Visual: 視覺動態特效
  • MPS: 系統 MetalPerformanceShaders.

最後

• 慢慢再補充其他相關濾鏡，喜歡就給我點個星🌟吧。
• 濾鏡Demo地址，目前包含100+種濾鏡，同時也支持CoreImage混合使用。
• 再附上一個開發加速庫KJCategoriesDemo地址
• 再附上一個網絡基礎庫RxNetworksDemo地址
• 喜歡的老闆們可以點個星🌟，謝謝各位老闆！！！

✌️.