引言

如果想在之上叠加一个静止图片很简单，像ffmpeg的滤镜、opencv等都能实现。但是假如文字拥有动画，而且文字出现比较频繁，全部使用序列的png图像会很大。例如如下的素材：

黑色素材

虽然与白色素材叠加算法中所用素材相同，但目的不同，以下demo将素材以“黑色部分叠加，白色区域透明”的效果叠加到视频之上。

算法实现

原视频：

input

#include <stdio.h>
#include <math.h>
#ifdef __cplusplus
extern "C" {
#endif
#include <libavformat/avformat.h>
#include <libavcodec/avcodec.h>
#include <libavutil/frame.h>
#include <libavutil/opt.h>
#include <libavutil/imgutils.h>
#include <libavutil/pixfmt.h>
#ifdef __cplusplus
};
#endif
int frame_cover_black( AVFrame* dst_frame, AVFrame* src_frame, AVFrame* cover_frame);
int init_frame(AVFrame* frame,int width,int height,uint8_t* dst_buff);
int mergeyuv(char* file, char* light,char* lightout,int width,int height);
int write_yuvframe(AVFrame *pFrame,FILE *out);
int main (char** args, int argv)
{
    char* file = "test.yuv";
    char* light = "light.yuv";
    char* lightout = "lightout.yuv";
    int width = 480;
    int height = 480;
    mergeyuv(file,light,lightout,width,height);
    return 0;
}
int mergeyuv(char* file, char* light,char* lightout,int width,int height)
{
    AVFrame* readframe,*lightframe,*outframe;
    readframe = av_frame_alloc();
    lightframe = av_frame_alloc();
    outframe = av_frame_alloc();
    FILE* readfile = (FILE*)fopen(file,"rb");
    FILE* lightfile = (FILE*)fopen(light,"rb");
    FILE* outfile = (FILE*)fopen(lightout,"wb+");
    if(lightfile==NULL||readframe==NULL||outfile==NULL)
        return -1;
    int length = width*height*3/2;
    uint8_t* readbuff = (uint8_t*)malloc(length);
    uint8_t* lightbuff = (uint8_t*)malloc(length);
    uint8_t* outbuff = (uint8_t*)malloc(length);
    init_frame(readframe,width,height,readbuff);
    init_frame(lightframe,width,height,lightbuff);
    init_frame(outframe,width,height,outbuff);
    while(fread(readbuff,1,length,readfile))
    {
        if(fread(lightbuff,1,length,lightfile))
        {
            puts("mrege one frame");
            frame_cover_black(readframe,readframe,lightframe);
            write_yuvframe(readframe,outfile);
        }
        else
            break;
    }
    fclose(readfile);
    fclose(lightfile);
    fclose(outfile);
    free(readbuff);
    free(lightbuff);
    free(outbuff);
    av_frame_free(&readframe);
    av_frame_free(&lightframe);
    av_frame_free(&outframe);
    return 0;
}
int init_frame(AVFrame* frame,int width,int height,uint8_t* dst_buff)
{
    if(!avpicture_fill((AVPicture *) frame, dst_buff, AV_PIX_FMT_YUV420P,width,height))
    {
        puts("init frame error");
       av_frame_free(&frame);
        return NULL;
    }
    frame->width=width;
    frame->height=height;
    frame->format = AV_PIX_FMT_YUV420P;
    return 0;
}
int frame_cover_black( AVFrame* dst_frame, AVFrame* src_frame, AVFrame* cover_frame)
{
    if(dst_frame == NULL || src_frame == NULL || cover_frame == NULL)
    {
        puts("frame_cover_black input or output frame is NULL");
        return -1;
    }
    int w2 = cover_frame->width;
    int h2 = cover_frame->height;
    int i = 0, j = 0;
    int a, a2;
    float rat;
    for(i = 0; i < h2; i++)
    {
        for(j = 0; j < w2; j++)
        {
            a2 = cover_frame->data[0][i * cover_frame->linesize[0] + j];
            a2 = a2 >= 0 ? a2 : 256 + a2;
            if(a2 <= 32)
            {
                dst_frame->data[1][(j >> 1) + (int)(i >> 1) * (dst_frame->linesize[1])] = cover_frame->data[1][(j >> 1) + (int)(i >> 1) * (cover_frame->linesize[1])];
                dst_frame->data[2][(j >> 1) + (int)(i >> 1) * (dst_frame->linesize[2])] = cover_frame->data[2][(j >> 1) + (int)(i >> 1) * (cover_frame->linesize[2])];
                dst_frame->data[0][i * dst_frame->linesize[0] + j] = 16;
                continue;
            }
            a = src_frame->data[0][i * src_frame->linesize[0] + j];
            a  = a >= 0 ? a : 256 + a;
            rat = (double)a2 / 256;
            rat = rat > 1 ? 1 : rat;
            dst_frame->data[0][i * dst_frame->linesize[0] + j] = a * rat;
        }
    }
    return 0;
}
int write_yuvframe(AVFrame *pFrame,FILE *out)
{
    int height = pFrame->height,width = pFrame->width;
    if(pFrame==NULL)
    {
        puts("error:write frame is null");
        return -1;
    }
    if(out == NULL)
    {
        puts("give write file is null");
        return -1;
    }
    int j = 0;
    for (j = 0; j < height; j++)
        fwrite(pFrame->data[0] + j * pFrame->linesize[0], 1, width, out);
    for (j = 0; j < height / 2; j++)
        fwrite(pFrame->data[1] + j * pFrame->linesize[1], 1, width / 2, out);
    for (j = 0; j < height / 2; j++)
        fwrite(pFrame->data[2] + j * pFrame->linesize[2], 1, width / 2,out);
    return 0;
}

这是效果：

output

注：

选用16作拐点的话，会出现大量泛白区域，所以选用32作为拐点来分离出黑色区域。但是同样会忽视某些细节。当然，很可能这些细节是由于编码的“有损”而产生的。使用200 作为全透明峰值。
y为素材视频对应点的Y值，d（包含uv）为输出帧的数据
if y< 32
d设置为黑色
else if y< 200
d按比例趋近黑色
else
忽视素材叠加，取原帧对应点数据
具体解释参看视频叠加算法-白色素材叠加

三待改进