内置过滤器

Question

我试过这个：

CIFilter *dodgeFilter = [CIFilter filterWithName:@"CIColorDodgeBlendMode"];

替换：

GPUImageDivideBlendFilter *divideBlendFilter = [[GPUImageDivideBlendFilter alloc] init];

但效果并不相同。

Answer 1

内置过滤器

您是否尝试过CIDivideBlendMode？

CIImage *img1 = [[CIImage alloc] initWithImage:[UIImage imageNamed:@"img1.jpg"]];
CIImage *img2 = [[CIImage alloc] initWithImage:[UIImage imageNamed:@"img2.jpg"]];

CIFilter *filterBuiltin = [CIFilter filterWithName:@"CIDivideBlendMode"
                                     keysAndValues:@"inputImage", img1,
                                                   @"inputBackgroundImage", img2, nil];
CIImage *outputImageBuiltin = [filterBuiltin outputImage];
UIImage *filteredImageBuiltin = [self imageWithCIImage:outputImageBuiltin];

自定义过滤器

我认为尝试基于现有的GPUImageFilter创建自定义CIFilter会很有趣，因为iOS8允许我们这样做。这应该允许将任何GPUImageFilter翻译为其CIFilter对应方。

在开始之前，值得查看What You Need to Know Before Writing a Custom Filter和Core Image Kernel Language Reference

我们首先编写自定义内核，它与GPUImageDivideBlendFilter着色器非常相似。一个例外是核心图像内核语言中似乎不支持的控制流部分，我们将使用*_branch1和*_branch2乘数来解决这个问题。

创建CIFilter很简单：

为您的Xcode项目创建一个新的ImageDivideBlendFilter.cikernel（您的自定义过滤器内核）文件：

kernel vec4 GPUImageDivideBlendFilter(sampler image1, sampler image2)
{
    float EPSILON = 1e-4;
    vec4 base = sample(image1, samplerCoord(image1));
    vec4 overlay = sample(image2, samplerCoord(image2));

    float ra1 = overlay.a * base.a + overlay.r * (1.0 - base.a) + base.r * (1.0 - overlay.a);
    float ra2 = (base.r * overlay.a * overlay.a) / overlay.r + overlay.r * (1.0 - base.a) + base.r * (1.0 - overlay.a);

    // https://developer.apple.com/library/mac/documentation/GraphicsImaging/Reference/CIKernelLangRef/ci_gslang_ext.html#//apple_ref/doc/uid/TP40004397-CH206-TPXREF101
    // "Other flow control statements (if, for, while, do while) are supported only when the loop condition can be inferred at the time the code compiles"
    float ra_branch2 = step(EPSILON, overlay.a) * step(base.r / overlay.r, base.a / overlay.a);
    float ra_branch1 = step(ra_branch2, 0.5);

    float ra = ra1 * ra_branch1 + ra2 * ra_branch2;


    float ga1 = overlay.a * base.a + overlay.g * (1.0 - base.a) + base.g * (1.0 - overlay.a);
    float ga2 = (base.g * overlay.a * overlay.a) / overlay.g + overlay.g * (1.0 - base.a) + base.g * (1.0 - overlay.a);

    float ga_branch2 = step(EPSILON, overlay.a) * step(base.g / overlay.g, base.a / overlay.a);
    float ga_branch1 = step(ga_branch2, 0.5);

    float ga = ga1 * ga_branch1 + ga2 * ga_branch2;


    float ba1 = overlay.a * base.a + overlay.b * (1.0 - base.a) + base.b * (1.0 - overlay.a);
    float ba2 = (base.b * overlay.a * overlay.a) / overlay.b + overlay.b * (1.0 - base.a) + base.b * (1.0 - overlay.a);

    float ba_branch2 = step(EPSILON, overlay.a) * step(base.b / overlay.b, base.a / overlay.a);
    float ba_branch1 = step(ba_branch2, 0.5);

    float ba = ba1 * ba_branch1 + ba2 * ba_branch2;

    return vec4(ra, ga, ba, 1.0);
}

添加过滤器的界面和实现

// ImageDivideBlendFilter.h
#import <CoreImage/CoreImage.h>

@interface ImageDivideBlendFilter : CIFilter

@end


// ImageDivideBlendFilter.m
#import "ImageDivideBlendFilter.h"

@interface ImageDivideBlendFilter()
{
    CIImage *_image1;
    CIImage *_image2;
}

@end

@implementation ImageDivideBlendFilter

static CIColorKernel *imageDivideBlendKernel = nil;

+ (void)initialize
{
    // This will load the kernel code which will compiled at run time. We do this just once to optimize performances
    if (!imageDivideBlendKernel)
    {
        NSBundle *bundle = [NSBundle bundleForClass:[self class]];
        NSString *code = [NSString stringWithContentsOfFile:[bundle pathForResource: @"ImageDivideBlendFilter" ofType: @"cikernel"] encoding:NSUTF8StringEncoding error:nil];
        NSArray *kernels = [CIColorKernel kernelsWithString:code];
        imageDivideBlendKernel = [kernels firstObject];
    }
}

- (CIImage *)outputImage
{
    return [imageDivideBlendKernel applyWithExtent:_image1.extent roiCallback:nil arguments:@[_image1, _image2]];
}

+ (CIFilter *)filterWithName: (NSString *)name
{
    CIFilter  *filter;
    filter = [[self alloc] init];
    return filter;
}

@end

我们已准备好在我们的应用程序中使用新创建的自定义过滤器

- (void)filterDemo
{
    CIImage *img1 = [[CIImage alloc] initWithImage:[UIImage imageNamed:@"img1.jpg"]];
    CIImage *img2 = [[CIImage alloc] initWithImage:[UIImage imageNamed:@"img2.jpg"]];

    [ImageDivideBlendFilter class]; // preload kernel, it speeds up loading the filter if used multiple times

    CIFilter *filterCustom = [CIFilter filterWithName:@"ImageDivideBlendFilter" keysAndValues:@"image1", img2, @"image2", img1, nil];
    CIImage *outputImageCustom = [filterCustom outputImage];

    UIImage *filteredImageCustom = [self imageWithCIImage:outputImageCustom];
}

- (UIImage *)imageWithCIImage:(CIImage *)ciimage
{
    CIContext *context = [CIContext contextWithOptions:nil];
    CGImageRef cgimg = [context createCGImage:ciimage fromRect:[ciimage extent]];
    UIImage *newImg = [UIImage imageWithCGImage:cgimg];

    CGImageRelease(cgimg);

    return newImg;
}

内置和自定义过滤器产生相同的结果。

编辑：Swift版本

我在Github https://github.com/tcamin/CustomCoreImageFilteringDemo上提供了一个示例项目，展示了如何在Swift中进行CIF过滤。

Answer 2

答案中的CIColorKernel代码不起作用;实际上，任何将多个采样器对象（图像）传递给内核的尝试都会失败。

Answer 3

此外 - 我知道这与问题没有关系，但我觉得应该指出这一点 - 内核代码有一个向后的东西，特别是与预乘相关的函数。在使用alpha通道独立于其余三个时，取消对多个采样器（或颜色）对象的预乘;当您拥有成品时，将它们与预乘相结合。如果您未在计算中更改或混合或以其他方式使用两个采样器（或颜色）对象，则不要这样做。

GPUImage中的GPUImageDivideBlendFilter，CIImage的相应过滤器是什么？

3 个答案:

内置过滤器

自定义过滤器

编辑：Swift版本