因为在做直播，最近突然有个想法，自己做做滤镜试试看，实验了很多方法，下面谈谈我从网上了解的结合自己的想法，给大家分享一下。

首先，说到滤镜肯定首先想到iOS上著名的开源库GPUImage，熟悉GPUImage都知道自定义一个采集端GPUImageVideoCamera再添加滤镜Target即可实现滤镜，但是因为我们直播采用七牛做的，我查看七牛文档，并没有发现可以自己定制相机的地方。so，只能考虑对CVPixelBufferRef处理了。

先上结果，采用coreImage处理：

CVPixelBufferRef->CIImage->CIFilter->CIImage->CVPixelBufferRef

_coreImageContext = [CIContext contextWithEAGLContext:self.openGLESContext options:options];
- (CVPixelBufferRef)coreImageHandle:(CVPixelBufferRef)pixelBuffer
{
        CFAbsoluteTime elapsedTime, startTime = CFAbsoluteTimeGetCurrent();
        CIImage *inputImage = [CIImage imageWithCVPixelBuffer:pixelBuffer];
        [_coreImageFilter setValue:inputImage forKey:kCIInputImageKey];
        CIImage *outputImage = [_coreImageFilter outputImage];
        elapsedTime = CFAbsoluteTimeGetCurrent() - startTime;
        NSLog(@"Core Image frame time: %f", elapsedTime * 1000.0);
        [_coreImageContext render:outputImage toCVPixelBuffer:pixelBuffer];
        return pixelBuffer;
}

其中_coreImageContext采用openGL ES 处理，也可变为CPU处理，但是考虑性能还是采用GPU处理吧，此时处理时间一张大概1ms，完全可以支撑24帧的画面

第一条思路：

对CVPixelBufferRef提取纹理，扔给GPUImage filter 处理，然后再拿到处理完的纹理变为CVPixelBufferRef，传给七牛。因为openGL基础不足，未成功，暂时放弃
自己仿着GPUImageVideoCamera 里面
- (void)captureOutput:(AVCaptureOutput *)captureOutput didOutputSampleBuffer:(CMSampleBufferRef)sampleBuffer fromConnection:(AVCaptureConnection *)connection
方法将PixelBuffer变为纹理，但是注意七牛代理里面回传回来的Buffer pixelFormat=BGRA，所以提取纹理CVPixelBufferGetPlaneCount(cameraFrame)只有一层（待续）

第二条思路：

直接处理RGBA，性能消耗太大，放弃
改方法为转为灰度算法 CPU用量50-80%。太特么烫了

CVPixelBufferLockBaseAddress(pixelBuffer, 0);
CFAbsoluteTime elapsedTime, startTime = CFAbsoluteTimeGetCurrent();
unsigned char *data = (unsigned char *)CVPixelBufferGetBaseAddress(pixelBuffer);
size_t bufferHeight = CVPixelBufferGetHeight(pixelBuffer);
size_t bufferWidth = CVPixelBufferGetWidth(pixelBuffer);
NSInteger myDataLength = bufferWidth * bufferHeight * 4;
for (int i = 0; i < myDataLength; i+=4)
{
        UInt8 r_pixel = data[i];
        UInt8 g_pixel = data[i+1];
        UInt8 b_pixel = data[i+2];
        //Gray = R*0.299 + G*0.587 + B*0.114
        int outputRed = (r_pixel * 0.299) + (g_pixel *0.587) + (b_pixel * 0.114);
        int outputGreen = (r_pixel * 0.299) + (g_pixel *0.587) + (b_pixel * 0.114);
        int outputBlue = (r_pixel * 0.299) + (g_pixel *0.587) + (b_pixel * 0.114);
        if(outputRed>255)outputRed=255;
        if(outputGreen>255)outputGreen=255;
        if(outputBlue>255)outputBlue=255;
        data[i] = outputRed;
        data[i+1] = outputGreen;
        data[i+2] = outputBlue;
}

elapsedTime = CFAbsoluteTimeGetCurrent() - startTime;
NSLog(@"CPU frame time: %f", elapsedTime * 1000.0);
CVPixelBufferUnlockBaseAddress(pixelBuffer, 0);

第三条思路:

就是上面结果那个啦~~