FFmpeg中AVFilter模块实践指南

转载

在做音视频相关的开发过程大体如下所示，对于其中的编码/解码，整个流程相对比较固定，使用ffmpeg可以很好的完成这部分的开发。对其中的帧数据处理（包括音频和视频数据）则相对要多样化一些，比如对视频做尺寸变换，进行音频音量均衡，直播中的美颜处理，多路流合成等等，这些都是属于流程中的帧数据处理。今天要介绍FFmpeg中的AVFilter模块进行帧数据处理的开发，AVFilter模块对帧数据处理进行了很好的抽象。AVFilter中的filter graph（滤波器图）概念非常适合帧数据处理中的多级滤波处理，同时对滤波器的接口进行了规定，后期添加一些自定义的滤波器也是很方便。网上关于AVFilter的介绍大多是基于ffmpeg的命令使用，基于代码实现的很少，最近项目中正好要使用到了AVFilter，写个小结，希望对有同样需求的小伙伴有帮助。

原始音视频–>解码–>帧数据处理–>编码–>输出音视频

1. 主要结构体和API介绍

// 对filters系统的整体管理 struct AVFilterGraph { AVFilterContext **filters; unsigned nb_filters; }

// 定义filter本身的能力，拥有的pads，回调函数接口定义 struct AVFilter { const char *name; const AVFilterPad *inputs; const AVFilterPad *outputs; }

// filter实例，管理filter与外部的联系 struct AVFilterContext { const AVFilter *filter; char *name; AVFilterPad *input_pads; AVFilterLink **inputs; unsigned nb_inputs AVFilterPad *output_pads; AVFilterLink **outputs; unsigned nb_outputs; struct AVFilterGraph *graph; }

// 定义两个filters之间的联接 struct AVFilterLink { AVFilterContext *src; AVFilterPad *srcpad; AVFilterContext *dst; AVFilterPad *dstpad; struct AVFilterGraph *graph; }

// 定义filter的输入/输出接口 struct AVFilterPad { const char *name; AVFrame *(*get_video_buffer)(AVFilterLink *link, int w, int h); AVFrame *(*get_audio_buffer)(AVFilterLink *link, int nb_samples); int (*filter_frame)(AVFilterLink *link, AVFrame *frame); int (*request_frame)(AVFilterLink *link); }

struct AVFilterInOut { char *name; AVFilterContext *filter_ctx; int pad_idx; struct AVFilterInOut *next; }

在AVFilter模块中定义了AVFilter结构，很个AVFilter都是具有独立功能的节点，如scale filter的作用就是进行图像尺寸变换，overlay filter的作用就是进行图像的叠加，这里需要重点提的是两个特别的filter，一个是buffer，一个是buffersink，滤波器buffer代表filter graph中的源头，原始数据就往这个filter节点输入的；而滤波器buffersink代表filter graph中的输出节点，处理完成的数据从这个filter节点输出。

// 获取FFmpeg中定义的filter，调用该方法前需要先调用avfilter_register_all();进行滤波器注册 AVFilter *avfilter_get_by_name(const char *name); // 往源滤波器buffer中输入待处理的数据 int av_buffersrc_add_frame(AVFilterContext *ctx, AVFrame *frame); // 从目的滤波器buffersink中输出处理完的数据 int av_buffersink_get_frame(AVFilterContext *ctx, AVFrame *frame); // 创建一个滤波器图filter graph AVFilterGraph *avfilter_graph_alloc(void); // 创建一个滤波器实例AVFilterContext，并添加到AVFilterGraph中 int avfilter_graph_create_filter(AVFilterContext **filt_ctx, const AVFilter *filt, const char *name, const char *args, void *opaque, AVFilterGraph *graph_ctx); // 连接两个滤波器节点 int avfilter_link(AVFilterContext *src, unsigned srcpad, AVFilterContext *dst, unsigned dstpad);

2. AVFilter主体框架流程

在利用AVFilter进行音视频数据处理前先将在进行的处理流程绘制出来，现在以FFmpeg filter官方文档中的一个例子为例进行说明。

[main] input --> split ---------------------> overlay --> output | ^ |[tmp] [flip]| +-----> crop --> vflip -------+

这个例子的处理流程如上所示，首先使用split滤波器将input流分成两路流（main和tmp），然后分别对两路流进行处理。对于tmp流，先经过crop滤波器进行裁剪处理，再经过flip滤波器进行垂直方向上的翻转操作，输出的结果命名为flip流。再将main流和flip流输入到overlay滤波器进行合成操作。上图的input就是上面提过的buffer源滤波器，output就是上面的提过的buffersink滤波器。上图中每个节点都是一个AVFilterContext，每个连线就是AVFliterLink。所有这些信息都统一由AVFilterGraph来管理。

3. 实例实现

extern "C" { #include #include #include #include #include #include #include } int main(int argc, char* argv) { int ret = 0; // input yuv FILE* inFile = NULL; const char* inFileName = "sintel_480x272_yuv420p.yuv"; fopen_s(&inFile, inFileName, "rb+"); if (!inFile) { printf("Fail to open file\n"); return -1; } int in_width = 480; int in_height = 272; // output yuv FILE* outFile = NULL; const char* outFileName = "out_crop_vfilter.yuv"; fopen_s(&outFile, outFileName, "wb"); if (!outFile) { printf("Fail to create file for output\n"); return -1; } avfilter_register_all(); AVFilterGraph* filter_graph = avfilter_graph_alloc(); if (!filter_graph) { printf("Fail to create filter graph!\n"); return -1; } // source filter char args[512]; _snprintf_s(args, sizeof(args), "video_size=%dx%d:pix_fmt=%d:time_base=%d/%d:pixel_aspect=%d/%d", in_width, in_height, AV_PIX_FMT_YUV420P, 1, 25, 1, 1); AVFilter* bufferSrc = avfilter_get_by_name("buffer"); AVFilterContext* bufferSrc_ctx; ret = avfilter_graph_create_filter(&bufferSrc_ctx, bufferSrc, "in", args, NULL, filter_graph); if (ret < 0) { printf("Fail to create filter bufferSrc\n"); return -1; } // sink filter AVBufferSinkParams *bufferSink_params; AVFilterContext* bufferSink_ctx; AVFilter* bufferSink = avfilter_get_by_name("buffersink"); enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE }; bufferSink_params = av_buffersink_params_alloc(); bufferSink_params->pixel_fmts = pix_fmts; ret = avfilter_graph_create_filter(&bufferSink_ctx, bufferSink, "out", NULL, bufferSink_params, filter_graph); if (ret < 0) { printf("Fail to create filter sink filter\n"); return -1; } // split filter AVFilter *splitFilter = avfilter_get_by_name("split"); AVFilterContext *splitFilter_ctx; ret = avfilter_graph_create_filter(&splitFilter_ctx, splitFilter, "split", "outputs=2", NULL, filter_graph); if (ret < 0) { printf("Fail to create split filter\n"); return -1; } // crop filter AVFilter *cropFilter = avfilter_get_by_name("crop"); AVFilterContext *cropFilter_ctx; ret = avfilter_graph_create_filter(&cropFilter_ctx, cropFilter, "crop", "out_w=iw:out_h=ih/2:x=0:y=0", NULL, filter_graph); if (ret < 0) { printf("Fail to create crop filter\n"); return -1; } // vflip filter AVFilter *vflipFilter = avfilter_get_by_name("vflip"); AVFilterContext *vflipFilter_ctx; ret = avfilter_graph_create_filter(&vflipFilter_ctx, vflipFilter, "vflip", NULL, NULL, filter_graph); if (ret < 0) { printf("Fail to create vflip filter\n"); return -1; } // overlay filter AVFilter *overlayFilter = avfilter_get_by_name("overlay"); AVFilterContext *overlayFilter_ctx; ret = avfilter_graph_create_filter(&overlayFilter_ctx, overlayFilter, "overlay", "y=0:H/2", NULL, filter_graph); if (ret < 0) { printf("Fail to create overlay filter\n"); return -1; } // src filter to split filter ret = avfilter_link(bufferSrc_ctx, 0, splitFilter_ctx, 0); if (ret != 0) { printf("Fail to link src filter and split filter\n"); return -1; } // split filter's first pad to overlay filter's main pad ret = avfilter_link(splitFilter_ctx, 0, overlayFilter_ctx, 0); if (ret != 0) { printf("Fail to link split filter and overlay filter main pad\n"); return -1; } // split filter's second pad to crop filter ret = avfilter_link(splitFilter_ctx, 1, cropFilter_ctx, 0); if (ret != 0) { printf("Fail to link split filter's second pad and crop filter\n"); return -1; } // crop filter to vflip filter ret = avfilter_link(cropFilter_ctx, 0, vflipFilter_ctx, 0); if (ret != 0) { printf("Fail to link crop filter and vflip filter\n"); return -1; } // vflip filter to overlay filter's second pad ret = avfilter_link(vflipFilter_ctx, 0, overlayFilter_ctx, 1); if (ret != 0) { printf("Fail to link vflip filter and overlay filter's second pad\n"); return -1; } // overlay filter to sink filter ret = avfilter_link(overlayFilter_ctx, 0, bufferSink_ctx, 0); if (ret != 0) { printf("Fail to link overlay filter and sink filter\n"); return -1; } // check filter graph ret = avfilter_graph_config(filter_graph, NULL); if (ret < 0) { printf("Fail in filter graph\n"); return -1; } char *graph_str = avfilter_graph_dump(filter_graph, NULL); FILE* graphFile = NULL; fopen_s(&graphFile, "graphFile.txt", "w"); fprintf(graphFile, "%s", graph_str); av_free(graph_str); AVFrame *frame_in = av_frame_alloc(); unsigned char *frame_buffer_in = (unsigned char *)av_malloc(av_image_get_buffer_size(AV_PIX_FMT_YUV420P, in_width, in_height, 1)); av_image_fill_arrays(frame_in->data, frame_in->linesize, frame_buffer_in, AV_PIX_FMT_YUV420P, in_width, in_height, 1); AVFrame *frame_out = av_frame_alloc(); unsigned char *frame_buffer_out = (unsigned char *)av_malloc(av_image_get_buffer_size(AV_PIX_FMT_YUV420P, in_width, in_height, 1)); av_image_fill_arrays(frame_out->data, frame_out->linesize, frame_buffer_out, AV_PIX_FMT_YUV420P, in_width, in_height, 1); frame_in->width = in_width; frame_in->height = in_height; frame_in->format = AV_PIX_FMT_YUV420P; while (1) { if (fread(frame_buffer_in, 1, in_width*in_height * 3 / 2, inFile) != in_width*in_height * 3 / 2) { break; } //input Y,U,V frame_in->data[0] = frame_buffer_in; frame_in->data[1] = frame_buffer_in + in_width*in_height; frame_in->data[2] = frame_buffer_in + in_width*in_height * 5 / 4; if (av_buffersrc_add_frame(bufferSrc_ctx, frame_in) < 0) { printf("Error while add frame.\n"); break; } /* pull filtered pictures from the filtergraph */ ret = av_buffersink_get_frame(bufferSink_ctx, frame_out); if (ret < 0) break; //output Y,U,V if (frame_out->format == AV_PIX_FMT_YUV420P) { for (int i = 0; i < frame_out->height; i++) { fwrite(frame_out->data[0] + frame_out->linesize[0] * i, 1, frame_out->width, outFile); } for (int i = 0; i < frame_out->height / 2; i++) { fwrite(frame_out->data[1] + frame_out->linesize[1] * i, 1, frame_out->width / 2, outFile); } for (int i = 0; i < frame_out->height / 2; i++) { fwrite(frame_out->data[2] + frame_out->linesize[2] * i, 1, frame_out->width / 2, outFile); } } printf("Process 1 frame!\n"); av_frame_unref(frame_out); } fclose(inFile); fclose(outFile); av_frame_free(&frame_in); av_frame_free(&frame_out); avfilter_graph_free(&filter_graph); return 0; }

github代码仓库

4. 建立定义Filter滤波器

这部分暂时还没有实践，参考ffmpeg源码中已有的filter和ffmpeg源码中的文档writing_filter.txt，应该实践起来也来难，等后面有时间再补上这部分介绍。