注：寫了一系列的結構體的分析的文章，在這裡列一個列表：

FFMPEG有幾個最重要的結構體，包含了解協議，解封裝，解碼操作，此前已經進行過分析：

在此不再詳述，其中AVFrame是包含碼流引數較多的結構體。本文將會詳細分析一下該結構體裡主要變數的含義和作用。

首先看一下結構體的定義（位於avcodec.h）：

/* *雷霄驊 *[email protected] *中國傳媒大學/數字電視技術 *//** * Audio Video Frame. * New fields can be added to the end of AVFRAME with minor version * bumps. Similarly fields that are marked as to be only accessed by * av_opt_ptr() can be reordered. This allows 2 forks to add fields * without breaking compatibility with each other. * Removal, reordering and changes in the remaining cases require * a major version bump. * sizeof(AVFrame) must not be used outside libavcodec. */
 
typedef struct AVFrame {#define AV_NUM_DATA_POINTERS 8    /**影象資料     * pointer to the picture/channel planes.     * This might be different from the first allocated byte     * - encoding: Set by user     * - decoding: set by AVCodecContext.get_buffer()     */    uint8_t *data[AV_NUM_DATA_POINTERS];    /**     * Size, in bytes, of the data for each picture/channel plane.     *     * For audio, only linesize[0] may be set. For planar audio, each channel     * plane must be the same size.     *     * - encoding: Set by user     * - decoding: set by AVCodecContext.get_buffer()     */
 
    int linesize[AV_NUM_DATA_POINTERS];    /**     * pointers to the data planes/channels.     *     * For video, this should simply point to data[].     *     * For planar audio, each channel has a separate data pointer, and     * linesize[0] contains the size of each channel buffer.     * For packed audio, there is just one data pointer, and linesize[0]     * contains the total size of the buffer for all channels.     *     * Note: Both data and extended_data will always be set by get_buffer(),     * but for planar audio with more channels that can fit in data,     * extended_data must be used by the decoder in order to access all     * channels.     *     * encoding: unused     * decoding: set by AVCodecContext.get_buffer()     */
 
    uint8_t **extended_data;    /**寬高     * width and height of the video frame     * - encoding: unused     * - decoding: Read by user.     */    int width, height;    /**     * number of audio samples (per channel) described by this frame     * - encoding: Set by user     * - decoding: Set by libavcodec     */    int nb_samples;    /**     * format of the frame, -1 if unknown or unset     * Values correspond to enum AVPixelFormat for video frames,     * enum AVSampleFormat for audio)     * - encoding: unused     * - decoding: Read by user.     */    int format;    /**是否是關鍵幀     * 1 -> keyframe, 0-> not     * - encoding: Set by libavcodec.     * - decoding: Set by libavcodec.     */    int key_frame;    /**幀型別（I,B,P）     * Picture type of the frame, see ?_TYPE below.     * - encoding: Set by libavcodec. for coded_picture (and set by user for input).     * - decoding: Set by libavcodec.     */    enum AVPictureType pict_type;    /**     * pointer to the first allocated byte of the picture. Can be used in get_buffer/release_buffer.     * This isn't used by libavcodec unless the default get/release_buffer() is used.     * - encoding:     * - decoding:     */    uint8_t *base[AV_NUM_DATA_POINTERS];    /**     * sample aspect ratio for the video frame, 0/1 if unknown/unspecified     * - encoding: unused     * - decoding: Read by user.     */    AVRational sample_aspect_ratio;    /**     * presentation timestamp in time_base units (time when frame should be shown to user)     * If AV_NOPTS_VALUE then frame_rate = 1/time_base will be assumed.     * - encoding: MUST be set by user.     * - decoding: Set by libavcodec.     */    int64_t pts;    /**     * reordered pts from the last AVPacket that has been input into the decoder     * - encoding: unused     * - decoding: Read by user.     */    int64_t pkt_pts;    /**     * dts from the last AVPacket that has been input into the decoder     * - encoding: unused     * - decoding: Read by user.     */    int64_t pkt_dts;    /**     * picture number in bitstream order     * - encoding: set by     * - decoding: Set by libavcodec.     */    int coded_picture_number;    /**     * picture number in display order     * - encoding: set by     * - decoding: Set by libavcodec.     */    int display_picture_number;    /**     * quality (between 1 (good) and FF_LAMBDA_MAX (bad))     * - encoding: Set by libavcodec. for coded_picture (and set by user for input).     * - decoding: Set by libavcodec.     */    int quality;    /**     * is this picture used as reference     * The values for this are the same as the MpegEncContext.picture_structure     * variable, that is 1->top field, 2->bottom field, 3->frame/both fields.     * Set to 4 for delayed, non-reference frames.     * - encoding: unused     * - decoding: Set by libavcodec. (before get_buffer() call)).     */    int reference;    /**QP表     * QP table     * - encoding: unused     * - decoding: Set by libavcodec.     */    int8_t *qscale_table;    /**     * QP store stride     * - encoding: unused     * - decoding: Set by libavcodec.     */    int qstride;    /**     *     */    int qscale_type;    /**跳過巨集塊表     * mbskip_table[mb]>=1 if MB didn't change     * stride= mb_width = (width+15)>>4     * - encoding: unused     * - decoding: Set by libavcodec.     */    uint8_t *mbskip_table;    /**運動矢量表     * motion vector table     * @code     * example:     * int mv_sample_log2= 4 - motion_subsample_log2;     * int mb_width= (width+15)>>4;     * int mv_stride= (mb_width << mv_sample_log2) + 1;     * motion_val[direction][x + y*mv_stride][0->mv_x, 1->mv_y];     * @endcode     * - encoding: Set by user.     * - decoding: Set by libavcodec.     */    int16_t (*motion_val[2])[2];    /**巨集塊型別表     * macroblock type table     * mb_type_base + mb_width + 2     * - encoding: Set by user.     * - decoding: Set by libavcodec.     */    uint32_t *mb_type;    /**DCT係數     * DCT coefficients     * - encoding: unused     * - decoding: Set by libavcodec.     */    short *dct_coeff;    /**參考幀列表     * motion reference frame index     * the order in which these are stored can depend on the codec.     * - encoding: Set by user.     * - decoding: Set by libavcodec.     */    int8_t *ref_index[2];    /**     * for some private data of the user     * - encoding: unused     * - decoding: Set by user.     */    void *opaque;    /**     * error     * - encoding: Set by libavcodec. if flags&CODEC_FLAG_PSNR.     * - decoding: unused     */    uint64_t error[AV_NUM_DATA_POINTERS];    /**     * type of the buffer (to keep track of who has to deallocate data[*])     * - encoding: Set by the one who allocates it.     * - decoding: Set by the one who allocates it.     * Note: User allocated (direct rendering) & internal buffers cannot coexist currently.     */    int type;    /**     * When decoding, this signals how much the picture must be delayed.     * extra_delay = repeat_pict / (2*fps)     * - encoding: unused     * - decoding: Set by libavcodec.     */    int repeat_pict;    /**     * The content of the picture is interlaced.     * - encoding: Set by user.     * - decoding: Set by libavcodec. (default 0)     */    int interlaced_frame;    /**     * If the content is interlaced, is top field displayed first.     * - encoding: Set by user.     * - decoding: Set by libavcodec.     */    int top_field_first;    /**     * Tell user application that palette has changed from previous frame.     * - encoding: ??? (no palette-enabled encoder yet)     * - decoding: Set by libavcodec. (default 0).     */    int palette_has_changed;    /**     * codec suggestion on buffer type if != 0     * - encoding: unused     * - decoding: Set by libavcodec. (before get_buffer() call)).     */    int buffer_hints;    /**     * Pan scan.     * - encoding: Set by user.     * - decoding: Set by libavcodec.     */    AVPanScan *pan_scan;    /**     * reordered opaque 64bit (generally an integer or a double precision float     * PTS but can be anything).     * The user sets AVCodecContext.reordered_opaque to represent the input at     * that time,     * the decoder reorders values as needed and sets AVFrame.reordered_opaque     * to exactly one of the values provided by the user through AVCodecContext.reordered_opaque     * @deprecated in favor of pkt_pts     * - encoding: unused     * - decoding: Read by user.     */    int64_t reordered_opaque;    /**     * hardware accelerator private data (FFmpeg-allocated)     * - encoding: unused     * - decoding: Set by libavcodec     */    void *hwaccel_picture_private;    /**     * the AVCodecContext which ff_thread_get_buffer() was last called on     * - encoding: Set by libavcodec.     * - decoding: Set by libavcodec.     */    struct AVCodecContext *owner;    /**     * used by multithreading to store frame-specific info     * - encoding: Set by libavcodec.     * - decoding: Set by libavcodec.     */    void *thread_opaque;    /**     * log2 of the size of the block which a single vector in motion_val represents:     * (4->16x16, 3->8x8, 2-> 4x4, 1-> 2x2)     * - encoding: unused     * - decoding: Set by libavcodec.     */    uint8_t motion_subsample_log2;    /**（音訊）取樣率     * Sample rate of the audio data.     *     * - encoding: unused     * - decoding: read by user     */    int sample_rate;    /**     * Channel layout of the audio data.     *     * - encoding: unused     * - decoding: read by user.     */    uint64_t channel_layout;    /**     * frame timestamp estimated using various heuristics, in stream time base     * Code outside libavcodec should access this field using:     * av_frame_get_best_effort_timestamp(frame)     * - encoding: unused     * - decoding: set by libavcodec, read by user.     */    int64_t best_effort_timestamp;    /**     * reordered pos from the last AVPacket that has been input into the decoder     * Code outside libavcodec should access this field using:     * av_frame_get_pkt_pos(frame)     * - encoding: unused     * - decoding: Read by user.     */    int64_t pkt_pos;    /**     * duration of the corresponding packet, expressed in     * AVStream->time_base units, 0 if unknown.     * Code outside libavcodec should access this field using:     * av_frame_get_pkt_duration(frame)     * - encoding: unused     * - decoding: Read by user.     */    int64_t pkt_duration;    /**     * metadata.     * Code outside libavcodec should access this field using:     * av_frame_get_metadata(frame)     * - encoding: Set by user.     * - decoding: Set by libavcodec.     */    AVDictionary *metadata;    /**     * decode error flags of the frame, set to a combination of     * FF_DECODE_ERROR_xxx flags if the decoder produced a frame, but there     * were errors during the decoding.     * Code outside libavcodec should access this field using:     * av_frame_get_decode_error_flags(frame)     * - encoding: unused     * - decoding: set by libavcodec, read by user.     */    int decode_error_flags;#define FF_DECODE_ERROR_INVALID_BITSTREAM   1#define FF_DECODE_ERROR_MISSING_REFERENCE   2    /**     * number of audio channels, only used for audio.     * Code outside libavcodec should access this field using:     * av_frame_get_channels(frame)     * - encoding: unused     * - decoding: Read by user.     */    int64_t channels;} AVFrame;

下面看幾個主要變數的作用（在這裡考慮解碼的情況）：

uint8_t *data[AV_NUM_DATA_POINTERS]：解碼後原始資料（對視訊來說是YUV，RGB，對音訊來說是PCM）

int linesize[AV_NUM_DATA_POINTERS]：data中“一行”資料的大小。注意：未必等於影象的寬，一般大於影象的寬。

int width, height：視訊幀寬和高（1920x1080,1280x720...）

int nb_samples：音訊的一個AVFrame中可能包含多個音訊幀，在此標記包含了幾個

int format：解碼後原始資料型別（YUV420，YUV422，RGB24...）

int key_frame：是否是關鍵幀

enum AVPictureType pict_type：幀型別（I,B,P...）

AVRational sample_aspect_ratio：寬高比（16:9，4:3...）

int64_t pts：顯示時間戳

int coded_picture_number：編碼幀序號

int display_picture_number：顯示幀序號

int8_t *qscale_table：QP表

uint8_t *mbskip_table：跳過巨集塊表

int16_t (*motion_val[2])[2]：運動矢量表

uint32_t *mb_type：巨集塊型別表

short *dct_coeff：DCT係數，這個沒有提取過

int8_t *ref_index[2]：運動估計參考幀列表（貌似H.264這種比較新的標準才會涉及到多參考幀）

int interlaced_frame：是否是隔行掃描

uint8_t motion_subsample_log2：一個巨集塊中的運動向量取樣個數，取log的

其他的變數不再一一列舉，原始碼中都有詳細的說明。在這裡重點分析一下幾個需要一定的理解的變數：

1.data[]

對於packed格式的資料（例如RGB24），會存到data[0]裡面。

對於planar格式的資料（例如YUV420P），則會分開成data[0]，data[1]，data[2]...（YUV420P中data[0]存Y，data[1]存U，data[2]存V）

2.pict_type

包含以下型別：

enum AVPictureType {    AV_PICTURE_TYPE_NONE = 0, ///< Undefined    AV_PICTURE_TYPE_I,     ///< Intra    AV_PICTURE_TYPE_P,     ///< Predicted    AV_PICTURE_TYPE_B,     ///< Bi-dir predicted    AV_PICTURE_TYPE_S,     ///< S(GMC)-VOP MPEG4    AV_PICTURE_TYPE_SI,    ///< Switching Intra    AV_PICTURE_TYPE_SP,    ///< Switching Predicted    AV_PICTURE_TYPE_BI,    ///< BI type};

寬高比是一個分數，FFMPEG中用AVRational表達分數：

/** * rational number numerator/denominator */typedef struct AVRational{    int num; ///< numerator    int den; ///< denominator} AVRational;

QP表指向一塊記憶體，裡面儲存的是每個巨集塊的QP值。巨集塊的標號是從左往右，一行一行的來的。每個巨集塊對應1個QP。

qscale_table[0]就是第1行第1列巨集塊的QP值；qscale_table[1]就是第1行第2列巨集塊的QP值；qscale_table[2]就是第1行第3列巨集塊的QP值。以此類推...

巨集塊的個數用下式計算：

注：巨集塊大小是16x16的。

每行巨集塊數：

int mb_stride = pCodecCtx->width/16+1

巨集塊的總數：

int mb_sum = ((pCodecCtx->height+15)>>4)*(pCodecCtx->width/16+1)

5.motion_subsample_log2

1個運動向量所能代表的畫面大小（用寬或者高表示，單位是畫素），注意，這裡取了log2。

程式碼註釋中給出以下資料：

4->16x16, 3->8x8, 2-> 4x4, 1-> 2x2

即1個運動向量代表16x16的畫面的時候，該值取4；1個運動向量代表8x8的畫面的時候，該值取3...以此類推

6.motion_val

運動矢量表儲存了一幀視訊中的所有運動向量。

該值的儲存方式比較特別：

int16_t (*motion_val[2])[2];

註釋中給了一段程式碼：

int mv_sample_log2= 4 - motion_subsample_log2;int mb_width= (width+15)>>4;int mv_stride= (mb_width << mv_sample_log2) + 1;motion_val[direction][x + y*mv_stride][0->mv_x, 1->mv_y];

大概知道了該資料的結構：

1.首先分為兩個列表L0和L1

2.每個列表（L0或L1）儲存了一系列的MV（每個MV對應一個畫面，大小由motion_subsample_log2決定）

3.每個MV分為橫座標和縱座標（x,y）

注意，在FFMPEG中MV和MB在儲存的結構上是沒有什麼關聯的，第1個MV是螢幕上左上角畫面的MV（畫面的大小取決於motion_subsample_log2），第2個MV是螢幕上第1行第2列的畫面的MV，以此類推。因此在一個巨集塊（16x16）的運動向量很有可能如下圖所示（line代表一行運動向量的個數）：

//例如8x8劃分的運動向量與巨集塊的關係：    //-------------------------    //|          |            |    //|mv[x]     |mv[x+1]     |    //-------------------------    //|          |           |    //|mv[x+line]|mv[x+line+1]|    //-------------------------

7.mb_type

巨集塊型別表儲存了一幀視訊中的所有巨集塊的型別。其儲存方式和QP表差不多。只不過其是uint32型別的，而QP表是uint8型別的。每個巨集塊對應一個巨集塊型別變數。

巨集塊型別如下定義所示：

//The following defines may change, don't expect compatibility if you use them.#define MB_TYPE_INTRA4x4   0x0001#define MB_TYPE_INTRA16x16 0x0002 //FIXME H.264-specific#define MB_TYPE_INTRA_PCM  0x0004 //FIXME H.264-specific#define MB_TYPE_16x16      0x0008#define MB_TYPE_16x8       0x0010#define MB_TYPE_8x16       0x0020#define MB_TYPE_8x8        0x0040#define MB_TYPE_INTERLACED 0x0080#define MB_TYPE_DIRECT2    0x0100 //FIXME#define MB_TYPE_ACPRED     0x0200#define MB_TYPE_GMC        0x0400#define MB_TYPE_SKIP       0x0800#define MB_TYPE_P0L0       0x1000#define MB_TYPE_P1L0       0x2000#define MB_TYPE_P0L1       0x4000#define MB_TYPE_P1L1       0x8000#define MB_TYPE_L0         (MB_TYPE_P0L0 | MB_TYPE_P1L0)#define MB_TYPE_L1         (MB_TYPE_P0L1 | MB_TYPE_P1L1)#define MB_TYPE_L0L1       (MB_TYPE_L0   | MB_TYPE_L1)#define MB_TYPE_QUANT      0x00010000#define MB_TYPE_CBP        0x00020000//Note bits 24-31 are reserved for codec specific use (h264 ref0, mpeg1 0mv, ...)

8.ref_index

運動估計參考幀列表儲存了一幀視訊中所有巨集塊的參考幀索引。這個列表其實在比較早的壓縮編碼標準中是沒有什麼用的。只有像H.264這樣的編碼標準才有多參考幀的概念。但是這個欄位目前我還沒有研究透。只是知道每個巨集塊包含有4個該值，該值反映的是參考幀的索引。以後有機會再進行細研究吧。

在這裡展示一下自己做的碼流分析軟體的執行結果。將上文介紹的幾個列表影象化顯示了出來（在這裡是使用MFC的繪圖函式畫出來的）

視訊幀：

QP引數提取的結果：

美化過的（加上了顏色）：

巨集塊型別引數提取的結果：

美化過的（加上了顏色，更清晰一些，s代表skip巨集塊）：

運動向量引數提取的結果（在這裡是List0）：

運動估計參考幀引數提取的結果：