使用ffmpeg生成視頻縮略圖

整個項目在 https://github.com/ximikang/ffmpegThumbnail發佈

生成縮略圖的步驟

1. 使用ffmpeg解碼視頻
2. 幀格式轉換
3. 根據縮略圖的數量從視頻流中取幀
4. 使用opencv創建畫布並生成縮略圖

ffmpeg解碼視頻

根據縮略圖的數量從視頻流中取幀

1. 獲取圖片之間的時間間隔

// Read media file and read the header information from container format
    AVFormatContext* pFormatContext = avformat_alloc_context();
    if (!pFormatContext) {
        logging("ERROR could not allocate memory for format context");
        return -1;
    }
    
    if (avformat_open_input(&pFormatContext, inputFilePath.string().c_str(), NULL, NULL) != 0) {
        logging("ERROR could not open media file");
    }
    
    logging("format %s, duration %lld us, bit_rate %lld", pFormatContext->iformat->name, pFormatContext->duration, pFormatContext->bit_rate);
    cout << "視頻時常：" << pFormatContext->duration / 1000.0 / 1000.0 << "s" << endl;
    int64_t video_duration = pFormatContext->duration;
    int sum_count = rowNums * colNums;
    //跳轉的間隔 ms
    int64_t time_step = video_duration / sum_count / 1000;

1. 設置跳轉時間獲取不一樣的視頻Packet

for (int i = 0; i < sum_count ; ++i) {
        cv::Mat tempImage;
        // 每次讀取相同時間間隔的圖像並存入vImage中
        while (av_read_frame(pFormatContext, pPacket) >= 0) {
            if (pPacket->stream_index == video_stream_index) {
                response = decode_packet_2mat(pPacket, pCodecContext, pFrame, tempImage);// 返回
            }
            if (response == 0)// 成功讀取一幀
                break;
            if (response < 0)
                continue;
        }
        vImage.push_back(tempImage);
        // 跳轉視頻
        av_seek_frame(pFormatContext, -1, ((double)time_step / (double)1000)* AV_TIME_BASE*(double)(i+1) + (double)pFormatContext->start_time, AVSEEK_FLAG_BACKWARD);
    }

3.獲取Frame
在固定的時間點可能沒法獲取從當前時間點的Packet獲取對應的Frame，因此須要對獲取的Packet進行判斷，若是沒有獲取到對應的Frame應該繼續獲取下一Packet直到獲取到對應的Frame爲止。

static int decode_packet_2mat(AVPacket* pPacket, AVCodecContext* pCodecContext, AVFrame* pFrame, cv::Mat& image) {
    int response = avcodec_send_packet(pCodecContext, pPacket);

    if (response < 0) {
        logging("Error while sending a packet to the decoder");
        return response;
    }

    while (response >= 0) {
        // return decoded out data from a decoder
        response = avcodec_receive_frame(pCodecContext, pFrame);
        if (response == AVERROR(EAGAIN) || response == AVERROR_EOF) {
            logging("averror averror_eof");
            break;
        }
        else if (response < 0) {
            logging("Error while receiving frame");
            return response;
        }

        if (response >= 0) {
            // 獲取到Frame
            image = frame2Mat(pFrame, pCodecContext->pix_fmt);
        } 
        return 0;
    }
}

幀格式轉換

因爲從視頻流獲取的幀是YUV格式的Frame格式，後面使用opencv進行操做因此進行格式轉換。

先使用ffmpeg中的SwsContext將從視頻中抽取到的幀從YUV轉換到BGR格式，再從BGRFrame中的內存中獲取原始數據，並轉換到opencv的Mat類型。

cv::Mat frame2Mat(AVFrame* pFrame, AVPixelFormat pPixFormat)
{
    // image init
    AVFrame* pRGBFrame = av_frame_alloc();
    uint8_t* out_buffer = new uint8_t[avpicture_get_size(AV_PIX_FMT_BGR24, pFrame->width, pFrame->height)];
    avpicture_fill((AVPicture*)pRGBFrame, out_buffer, AV_PIX_FMT_BGR24, pFrame->width, pFrame->height);
    SwsContext* rgbSwsContext = sws_getContext(pFrame->width, pFrame->height, pPixFormat, pFrame->width, pFrame->height, AV_PIX_FMT_BGR24,SWS_BICUBIC, NULL, NULL, NULL);
    if (!rgbSwsContext) {
        logging("Error could not create frame to rgbframe sws context");
        exit(-1);
    }
    if (sws_scale(rgbSwsContext, pFrame->data, pFrame->linesize, 0, pFrame->height, pRGBFrame->data, pRGBFrame->linesize) < 0) {
        logging("Error could not sws to rgb frame");
        exit(-1);
    }

    cv::Mat mRGB(cv::Size(pFrame->width, pFrame->height), CV_8UC3);
    mRGB.data = (uchar*)pRGBFrame->data[0];//注意不能寫爲：(uchar*)pFrameBGR->data

    av_free(pRGBFrame);
    sws_freeContext(rgbSwsContext);
    return mRGB;
}

使用opencv創建畫布並生成縮略圖

經過畫布須要的大小參數，畫出白色畫布，再對畫布進行填充。

cv::Mat makeThumbnail(vector<cv::Mat> vImage, const unsigned int rowNums, const unsigned int colNums)
{
    // 判斷圖片時候知足條件
    if (vImage.size() != rowNums * colNums) {
        logging("Error image size not equal input size");
        logging("vImage length: %d, rowNums: %d, col number: %d", vImage.size(), rowNums, colNums);
        exit(-1);
    }
    int interval = 100;
    int height = vImage[0].size().height * rowNums + interval * (rowNums + 1);
    int width = vImage[0].size().width * colNums + interval * (colNums + 1);
    logging("thumbnail size: %d * %d", height, width);
    cv::Mat thumbnail(cv::Size(width, height), CV_8UC3);
    thumbnail.setTo(255);

    // 進行填充
    for (int i = 0; i < rowNums; ++i) {
        for (int j = 0; j < colNums; ++j) {
            int no = i * rowNums + j;
            int widthOffset = (vImage[0].size().width + interval) * j + interval;
            int heightOffset = (vImage[0].size().height + interval) * i + interval;
            vImage[no].copyTo(thumbnail(cv::Rect(widthOffset, heightOffset, vImage[0].size().width, vImage[0].size().height)));
        }
    }
    return thumbnail;
}

最後的效果