节选自一步一步解析H.264码流的NALU(SPS,PSS,IDR)
我看编/解码有个习惯就是从编/解码出来的数据文件下手。因为无论如何复杂,最终的文件(字节流)肯定是符合某种规范的。
这里我查看码流用的工具是StreamEye,码流文件是foreman_part_qcif.yuv编码得来的。这些工具怎么用就不介绍了,无非就是看视频有哪些/类帧组成,各个属性(Header/MacroBlock/Picture)是什么。
下面这段信息是从Headers Info拷贝出来的,粗看下它是SPS,PPS和Slice Header,那这有什么用,这些数据是什么意思?
[00]seq_parameter_set_rbsp() {
profile_idc = 66 (Baseline)
constraint_set0_flag = 0 (false)
constraint_set1_flag = 0 (false)
constraint_set2_flag = 0 (false)
constraint_set3_flag = 0 (false)
constraint_set4_flag = 0 (false)
constraint_set5_flag = 0 (false)
reserved_zero_2bits = 0
level_idc = 30
seq_parameter_set_id = 0
if (profile_idc == 100 || profile_idc == 110 ||
profile_idc == 122 || profile_idc == 144) {
chroma_format_idc = na
if (chroma_format_idc == 3)
separate_colour_plane_flag = na
bit_depth_luma_minus8 = na
bit_depth_chroma_minus8 = na
qpprime_y_zero_transform_bypass_flag = na
seq_scaling_matrix_present_flag = na
if (seq_scaling_matrix_present_flag)
for (i = 0; i < 8; i++) {
seq_scaling_list_present_flag[0] = na
if (seq_scaling_list_present_flag[0])
scaling_list_4x4[00] = na
scaling_list_4x4[01] = na
scaling_list_4x4[02] = na
scaling_list_4x4[03] = na
scaling_list_4x4[04] = na
scaling_list_4x4[05] = na
scaling_list_4x4[06] = na
scaling_list_4x4[07] = na
scaling_list_4x4[08] = na
scaling_list_4x4[09] = na
scaling_list_4x4[10] = na
scaling_list_4x4[11] = na
scaling_list_4x4[12] = na
scaling_list_4x4[13] = na
scaling_list_4x4[14] = na
scaling_list_4x4[15] = na
seq_scaling_list_present_flag[1] = na
if (seq_scaling_list_present_flag[1])
scaling_list_4x4[00] = na
scaling_list_4x4[01] = na
scaling_list_4x4[02] = na
scaling_list_4x4[03] = na
scaling_list_4x4[04] = na
scaling_list_4x4[05] = na
scaling_list_4x4[06] = na
scaling_list_4x4[07] = na
scaling_list_4x4[08] = na
scaling_list_4x4[09] = na
scaling_list_4x4[10] = na
scaling_list_4x4[11] = na
scaling_list_4x4[12] = na
scaling_list_4x4[13] = na
scaling_list_4x4[14] = na
scaling_list_4x4[15] = na
seq_scaling_list_present_flag[2] = na
if (seq_scaling_list_present_flag[2])
scaling_list_4x4[00] = na
scaling_list_4x4[01] = na
scaling_list_4x4[02] = na
scaling_list_4x4[03] = na
scaling_list_4x4[04] = na
scaling_list_4x4[05] = na
scaling_list_4x4[06] = na
scaling_list_4x4[07] = na
scaling_list_4x4[08] = na
scaling_list_4x4[09] = na
scaling_list_4x4[10] = na
scaling_list_4x4[11] = na
scaling_list_4x4[12] = na
scaling_list_4x4[13] = na
scaling_list_4x4[14] = na
scaling_list_4x4[15] = na
seq_scaling_list_present_flag[3] = na
if (seq_scaling_list_present_flag[3])
scaling_list_4x4[00] = na
scaling_list_4x4[01] = na
scaling_list_4x4[02] = na
scaling_list_4x4[03] = na
scaling_list_4x4[04] = na
scaling_list_4x4[05] = na
scaling_list_4x4[06] = na
scaling_list_4x4[07] = na
scaling_list_4x4[08] = na
scaling_list_4x4[09] = na
scaling_list_4x4[10] = na
scaling_list_4x4[11] = na
scaling_list_4x4[12] = na
scaling_list_4x4[13] = na
scaling_list_4x4[14] = na
scaling_list_4x4[15] = na
seq_scaling_list_present_flag[4] = na
if (seq_scaling_list_present_flag[4])
scaling_list_4x4[00] = na
scaling_list_4x4[01] = na
scaling_list_4x4[02] = na
scaling_list_4x4[03] = na
scaling_list_4x4[04] = na
scaling_list_4x4[05] = na
scaling_list_4x4[06] = na
scaling_list_4x4[07] = na
scaling_list_4x4[08] = na
scaling_list_4x4[09] = na
scaling_list_4x4[10] = na
scaling_list_4x4[11] = na
scaling_list_4x4[12] = na
scaling_list_4x4[13] = na
scaling_list_4x4[14] = na
scaling_list_4x4[15] = na
seq_scaling_list_present_flag[5] = na
if (seq_scaling_list_present_flag[5])
scaling_list_4x4[00] = na
scaling_list_4x4[01] = na
scaling_list_4x4[02] = na
scaling_list_4x4[03] = na
scaling_list_4x4[04] = na
scaling_list_4x4[05] = na
scaling_list_4x4[06] = na
scaling_list_4x4[07] = na
scaling_list_4x4[08] = na
scaling_list_4x4[09] = na
scaling_list_4x4[10] = na
scaling_list_4x4[11] = na
scaling_list_4x4[12] = na
scaling_list_4x4[13] = na
scaling_list_4x4[14] = na
scaling_list_4x4[15] = na
seq_scaling_list_present_flag[6] = na
if (seq_scaling_list_present_flag[6])
scaling_list_8x8[00] = na
scaling_list_8x8[01] = na
scaling_list_8x8[02] = na
scaling_list_8x8[03] = na
scaling_list_8x8[04] = na
scaling_list_8x8[05] = na
scaling_list_8x8[06] = na
scaling_list_8x8[07] = na
scaling_list_8x8[08] = na
scaling_list_8x8[09] = na
scaling_list_8x8[10] = na
scaling_list_8x8[11] = na
scaling_list_8x8[12] = na
scaling_list_8x8[13] = na
scaling_list_8x8[14] = na
scaling_list_8x8[15] = na
scaling_list_8x8[16] = na
scaling_list_8x8[17] = na
scaling_list_8x8[18] = na
scaling_list_8x8[19] = na
scaling_list_8x8[20] = na
scaling_list_8x8[21] = na
scaling_list_8x8[22] = na
scaling_list_8x8[23] = na
scaling_list_8x8[24] = na
scaling_list_8x8[25] = na
scaling_list_8x8[26] = na
scaling_list_8x8[27] = na
scaling_list_8x8[28] = na
scaling_list_8x8[29] = na
scaling_list_8x8[30] = na
scaling_list_8x8[31] = na
scaling_list_8x8[32] = na
scaling_list_8x8[33] = na
scaling_list_8x8[34] = na
scaling_list_8x8[35] = na
scaling_list_8x8[36] = na
scaling_list_8x8[37] = na
scaling_list_8x8[38] = na
scaling_list_8x8[39] = na
scaling_list_8x8[40] = na
scaling_list_8x8[41] = na
scaling_list_8x8[42] = na
scaling_list_8x8[43] = na
scaling_list_8x8[44] = na
scaling_list_8x8[45] = na
scaling_list_8x8[46] = na
scaling_list_8x8[47] = na
scaling_list_8x8[48] = na
scaling_list_8x8[49] = na
scaling_list_8x8[50] = na
scaling_list_8x8[51] = na
scaling_list_8x8[52] = na
scaling_list_8x8[53] = na
scaling_list_8x8[54] = na
scaling_list_8x8[55] = na
scaling_list_8x8[56] = na
scaling_list_8x8[57] = na
scaling_list_8x8[58] = na
scaling_list_8x8[59] = na
scaling_list_8x8[60] = na
scaling_list_8x8[61] = na
scaling_list_8x8[62] = na
scaling_list_8x8[63] = na
seq_scaling_list_present_flag[7] = na
if (seq_scaling_list_present_flag[7])
scaling_list_8x8[00] = na
scaling_list_8x8[01] = na
scaling_list_8x8[02] = na
scaling_list_8x8[03] = na
scaling_list_8x8[04] = na
scaling_list_8x8[05] = na
scaling_list_8x8[06] = na
scaling_list_8x8[07] = na
scaling_list_8x8[08] = na
scaling_list_8x8[09] = na
scaling_list_8x8[10] = na
scaling_list_8x8[11] = na
scaling_list_8x8[12] = na
scaling_list_8x8[13] = na
scaling_list_8x8[14] = na
scaling_list_8x8[15] = na
scaling_list_8x8[16] = na
scaling_list_8x8[17] = na
scaling_list_8x8[18] = na
scaling_list_8x8[19] = na
scaling_list_8x8[20] = na
scaling_list_8x8[21] = na
scaling_list_8x8[22] = na
scaling_list_8x8[23] = na
scaling_list_8x8[24] = na
scaling_list_8x8[25] = na
scaling_list_8x8[26] = na
scaling_list_8x8[27] = na
scaling_list_8x8[28] = na
scaling_list_8x8[29] = na
scaling_list_8x8[30] = na
scaling_list_8x8[31] = na
scaling_list_8x8[32] = na
scaling_list_8x8[33] = na
scaling_list_8x8[34] = na
scaling_list_8x8[35] = na
scaling_list_8x8[36] = na
scaling_list_8x8[37] = na
scaling_list_8x8[38] = na
scaling_list_8x8[39] = na
scaling_list_8x8[40] = na
scaling_list_8x8[41] = na
scaling_list_8x8[42] = na
scaling_list_8x8[43] = na
scaling_list_8x8[44] = na
scaling_list_8x8[45] = na
scaling_list_8x8[46] = na
scaling_list_8x8[47] = na
scaling_list_8x8[48] = na
scaling_list_8x8[49] = na
scaling_list_8x8[50] = na
scaling_list_8x8[51] = na
scaling_list_8x8[52] = na
scaling_list_8x8[53] = na
scaling_list_8x8[54] = na
scaling_list_8x8[55] = na
scaling_list_8x8[56] = na
scaling_list_8x8[57] = na
scaling_list_8x8[58] = na
scaling_list_8x8[59] = na
scaling_list_8x8[60] = na
scaling_list_8x8[61] = na
scaling_list_8x8[62] = na
scaling_list_8x8[63] = na
}
}
}
log2_max_frame_num_minus4 = 0 (4)
pic_order_cnt_type = 0
if (pic_order_cnt_type == 0)
log2_max_pic_order_cnt_lsb_minus4 = 0 (4)
else if (pic_order_cnt_type == 1) {
delta_pic_order_always_zero_flag = na
offset_for_non_ref_pic = na
offset_for_top_to_bottom_field = na
num_ref_frames_in_pic_order_cnt_cycle = na
for(i = 0; i < num_ref_frames_in_pic_order_cnt_cycle; i++)
}
max_num_ref_frames = 10
gaps_in_frame_num_value_allowed_flag = 0
pic_width_in_mbs_minus1 = 10 (176)
pic_height_in_map_units_minus1 = 8 (144)
frame_mbs_only_flag = 1
if (!frame_mbs_only_flag)
mb_adaptive_frame_field_flag = na
direct_8x8_inference_flag = 0 (false)
frame_cropping_flag = 0 (false)
if (frame_cropping_flag) {
frame_crop_left_offset = na
frame_crop_right_offset = na
frame_crop_top_offset = na
frame_crop_bottom_offset = na
}
vui_parameters_present_flag = 0 (false)
if (vui_parameters_present_flag)
}
vui_parameters()
}
[00]pic_parameter_set_rbsp() {
pic_parameter_set_id = 0
seq_parameter_set_id = 0
entropy_coding_mode_flag = 0 (CAVLC)
pic_order_present_flag = 0 (false)
num_slice_groups_minus1 = 0 (1)
if (num_slice_groups_minus1 > 0) {
slice_group_map_type = na
if (slice_group_map_type == 0)
for (iGroup = 0; iGroup <= num_slice_groups_minus1; iGroup++) {
}
else if (slice_group_map_type == 2)
for (iGroup = 0; iGroup < num_slice_groups_minus1; iGroup++) {
}
else if ((slice_group_map_type == 3) ||
(slice_group_map_type == 4) || (slice_group_map_type == 5)) {
slice_group_change_direction_flag = na
slice_group_change_rate_minus1 = na
} else if (slice_group_map_type == 6) {
pic_size_in_map_units_minus1 = na
for (i = 0; i <= pic_size_in_map_units_minus1; i++)
}
}
num_ref_idx_l0_active_minus1 = 9 (10)
num_ref_idx_l1_active_minus1 = 9 (10)
weighted_pred_flag = 0 (false)
weighted_bipred_idc = 0
pic_init_qp_minus26 = 0 (26)
pic_init_qs_minus26 = 0 (26)
chroma_qp_index_offset = 0
deblocking_filter_control_present_flag = 0 (false)
constrained_intra_pred_flag = 0 (false)
redundant_pic_cnt_present_flag = 0 (false)
}
if (more_rbsp_data()) {
transform_8x8_mode_flag = na
pic_scaling_matrix_present_flag = na
if (pic_scaling_matrix_present_flag) {
for (i = 0; i < 6 + 2 * transform_8x8_mode_flag; i++) {
pic_scaling_list_present_flag[0] = na
if (pic_scaling_list_present_flag[0])
scaling_list_4x4[0][00] = na
scaling_list_4x4[0][01] = na
scaling_list_4x4[0][02] = na
scaling_list_4x4[0][03] = na
scaling_list_4x4[0][04] = na
scaling_list_4x4[0][05] = na
scaling_list_4x4[0][06] = na
scaling_list_4x4[0][07] = na
scaling_list_4x4[0][08] = na
scaling_list_4x4[0][09] = na
scaling_list_4x4[0][10] = na
scaling_list_4x4[0][11] = na
scaling_list_4x4[0][12] = na
scaling_list_4x4[0][13] = na
scaling_list_4x4[0][14] = na
scaling_list_4x4[0][15] = na
pic_scaling_list_present_flag[1] = na
if (pic_scaling_list_present_flag[1])
scaling_list_4x4[1][00] = na
scaling_list_4x4[1][01] = na
scaling_list_4x4[1][02] = na
scaling_list_4x4[1][03] = na
scaling_list_4x4[1][04] = na
scaling_list_4x4[1][05] = na
scaling_list_4x4[1][06] = na
scaling_list_4x4[1][07] = na
scaling_list_4x4[1][08] = na
scaling_list_4x4[1][09] = na
scaling_list_4x4[1][10] = na
scaling_list_4x4[1][11] = na
scaling_list_4x4[1][12] = na
scaling_list_4x4[1][13] = na
scaling_list_4x4[1][14] = na
scaling_list_4x4[1][15] = na
pic_scaling_list_present_flag[2] = na
if (pic_scaling_list_present_flag[2])
scaling_list_4x4[2][00] = na
scaling_list_4x4[2][01] = na
scaling_list_4x4[2][02] = na
scaling_list_4x4[2][03] = na
scaling_list_4x4[2][04] = na
scaling_list_4x4[2][05] = na
scaling_list_4x4[2][06] = na
scaling_list_4x4[2][07] = na
scaling_list_4x4[2][08] = na
scaling_list_4x4[2][09] = na
scaling_list_4x4[2][10] = na
scaling_list_4x4[2][11] = na
scaling_list_4x4[2][12] = na
scaling_list_4x4[2][13] = na
scaling_list_4x4[2][14] = na
scaling_list_4x4[2][15] = na
pic_scaling_list_present_flag[3] = na
if (pic_scaling_list_present_flag[3])
scaling_list_4x4[3][00] = na
scaling_list_4x4[3][01] = na
scaling_list_4x4[3][02] = na
scaling_list_4x4[3][03] = na
scaling_list_4x4[3][04] = na
scaling_list_4x4[3][05] = na
scaling_list_4x4[3][06] = na
scaling_list_4x4[3][07] = na
scaling_list_4x4[3][08] = na
scaling_list_4x4[3][09] = na
scaling_list_4x4[3][10] = na
scaling_list_4x4[3][11] = na
scaling_list_4x4[3][12] = na
scaling_list_4x4[3][13] = na
scaling_list_4x4[3][14] = na
scaling_list_4x4[3][15] = na
pic_scaling_list_present_flag[4] = na
if (pic_scaling_list_present_flag[4])
scaling_list_4x4[4][00] = na
scaling_list_4x4[4][01] = na
scaling_list_4x4[4][02] = na
scaling_list_4x4[4][03] = na
scaling_list_4x4[4][04] = na
scaling_list_4x4[4][05] = na
scaling_list_4x4[4][06] = na
scaling_list_4x4[4][07] = na
scaling_list_4x4[4][08] = na
scaling_list_4x4[4][09] = na
scaling_list_4x4[4][10] = na
scaling_list_4x4[4][11] = na
scaling_list_4x4[4][12] = na
scaling_list_4x4[4][13] = na
scaling_list_4x4[4][14] = na
scaling_list_4x4[4][15] = na
pic_scaling_list_present_flag[5] = na
if (pic_scaling_list_present_flag[5])
scaling_list_4x4[5][00] = na
scaling_list_4x4[5][01] = na
scaling_list_4x4[5][02] = na
scaling_list_4x4[5][03] = na
scaling_list_4x4[5][04] = na
scaling_list_4x4[5][05] = na
scaling_list_4x4[5][06] = na
scaling_list_4x4[5][07] = na
scaling_list_4x4[5][08] = na
scaling_list_4x4[5][09] = na
scaling_list_4x4[5][10] = na
scaling_list_4x4[5][11] = na
scaling_list_4x4[5][12] = na
scaling_list_4x4[5][13] = na
scaling_list_4x4[5][14] = na
scaling_list_4x4[5][15] = na
}
}
second_chroma_qp_index_offset = na
}
}
[00]slice_header() {
nal_unit_header_svc_extension() {
idr_flag = na
priority_id = na
no_inter_layer_pred_flag = na
dependency_id = na
quality_id = na
temporal_id = na
use_ref_base_pic_flag = na
discardable_flag = na
output_flag = na
}
first_mb_in_slice = 0
slice_type = 7 (I slice)
pic_parameter_set_id = 0
frame_num = 0
if (!frame_mbs_only_flag) {
field_pic_flag = na
if (field_pic_flag)
bottom_field_flag = na
}
if (nal_unit_type == 5)
idr_pic_id = 0
if (pic_order_cnt_type == 0) {
pic_order_cnt_lsb = 0
if (pic_order_present_flag && !field_pic_flag)
delta_pic_order_cnt_bottom = na
}
if (pic_order_cnt_type == 1 && !delta_pic_order_always_zero_flag) {
delta_pic_order_cnt[0] = na
if (pic_order_present_flag && !field_pic_flag)
delta_pic_order_cnt[1] = na
}
if (redundant_pic_cnt_present_flag)
redundant_pic_cnt = na
if (slice_type == B)
direct_spatial_mv_pred_flag = na
if (slice_type == P || slice_type == SP || slice_type == B) {
num_ref_idx_active_override_flag = na
if (num_ref_idx_active_override_flag) {
num_ref_idx_l0_active_minus1 = na
if (slice_type == B )
num_ref_idx_l1_active_minus1 = na
}
}
if (nal_unit_type == 20)
ref_pic_list_mvc_modification()
else
ref_pic_list_modification()
if ((weighted_pred_flag && (slice_type == P ||
slice_type == SP)) || (weighted_bipred_idc == 1 && slice_type == B))
pred_weight_table()
if (nal_ref_idc != 0)
dec_ref_pic_marking()
if (entropy_coding_mode_flag && slice_type != I && slice_type != SI)
cabac_init_idc = na
slice_qp_delta = 2
if (slice_type == SP || slice_type == SI) {
if (slice_type == SP)
sp_for_switch_flag = na
slice_qs_delta = na
}
if (deblocking_filter_control_present_flag) {
disable_deblocking_filter_idc = na
if (disable_deblocking_filter_idc != 1) {
slice_alpha_c0_offset_div2 = na
slice_beta_offset_div2 = na
}
}
if (num_slice_groups_minus1 > 0 &&
slice_group_map_type >= 3 && slice_group_map_type <= 5)
slice_group_change_cycle = na
}
我们现在来分析,我们知道码流是由一个个的NAL Unit组成的,NALU是由NALU头和RBSP数据组成,而RBSP可能是SPS,PPS,Slice或SEI,目前我们这里SEI不会出现,而且SPS位于第一个NALU,PPS位于第二个NALU,其他就是Slice(严谨点区分的话可以把IDR等等再分出来)了。foreman_part_qcif.yuv只有3帧,那这里编码出来是不是就有5个NALU?我们这里可以大胆假设,然后仔细验证。
现在我们来开始分析,下面是一段H.264码流文件的十六进制数据,所以你得有个十六进制编辑器。00 00 00 01 67 42 00 1E F1 61 62 62 00 00 00 01 68 C8 A1 43 88 00
我们知道00 00 00 01
是NALU的开始标记,所以你打开这个完整的码流文件应该可以看到5个00 00 00 01
,所以这就是我们之前说的有5个NALU,分别是SPS,PPS和3个Slice。
一、SPS
这里SPS结构可以参考H264系列十五 句法元素SPS
去除掉NALU开始标记后的67 42 00 1E F1 61 62 62
,转换为二进制是01100111 01000010 00000000 00011110 11110001 01100001 01100010 01100010
。
1.先看NALU头
forbidden_zero_bit ,nal_ref_idc,nal_unit_type这三个属性共占8位(Spec上都有写,分别占1,2和5位),那我们对着解析下就看出
forbidden_zero_bit = 0 // 0
nal_ref_idc = 3 // 11
nal_unit_type = 7 // 00111
这就对了,看看#define NALU_TYPE_SPS 7
Spec当中后面有些放在if判断里的就是只有符合某个值的时候才会出现,我们这里nal_unit_type为7,不符合,所以直接跳过,进入到RBSP当中,这里是SPS,所以对照Spec
profile_idc
constraint_set0_flag
constraint_set1_flag
constraint_set2_flag
constraint_set3_flag
constraint_set4_flag
constraint_set5_flag
reserved_zero_2bits
level_idc
seq_parameter_set_id
这几个属性,直到seq_parameter_set_id之前都还比较好解析,我们就直接写出它们的值了
profile_idc = 66 // 01000010
constraint_set0_flag = 0 // 0
constraint_set1_flag = 0 // 0
constraint_set2_flag = 0 // 0
constraint_set3_flag = 0 // 0
constraint_set4_flag = 0 // 0
constraint_set5_flag = 0 // 0
reserved_zero_2bits = 0 // 00
level_idc = 30 // 00011110
对于seq_parameter_set_id,我们看到它是ue(v),这是一种Exp-Golomb编码,每个编码所占的位数不是固定的,我们现在还剩下的数据是11110001 01100001 01100010 01100010
。
2.指数哥伦布编码
关于指数哥伦布编码,可以参考H264系列九 热力学熵 信息熵 哈夫曼编码 哥伦布编码,这里节选如下
对数字20进行编码:
codeNum = 20
prefixLen = floor[log2(codeNum+1)] = floor[log2(21)]=4
surfix = codeNum+1-2^prefixLen=20+1-2^4=5=二进制的101
编码值=0000,1,0101(为方便观看,以逗号分隔了)
也就是说,哥伦布编码以中间的1为对称轴,前缀全写0,需要先算出一共要写几个0。然后再算后缀的信息位。至于解码,先计算前缀个数4个0,这样后缀的信息位是0101,也就是5。所以解码得出2^4-1+5=20
3.公式参考Spec(9.1 Parsing process for Exp-Golomb codes)
leadingZeroBits = −1
for (b = 0; !b; leadingZeroBits++)
b = read_bits(1)
codeNum = 2^(leadingZeroBits) − 1 + read_bits(leadingZeroBits)
对照上述解释,leadingZeroBits就是在找前缀个数,codeNum自然是我们最终解码值。现在还剩下的数据是11110001 01100001 01100010 01100010
这里read_bits直接读取到1了,说明前缀个数是0,跳出for循环,同时leadingZeroBits=0,后缀信息位不用读了,也是0。最终codeNum=2^0-1+0=0。也就是说这里编码1,解析出来的是seq_parameter_set_id = 0 // Exp-Golomb解1
。后面几个属性类似,都是指数哥伦布解析1,结果都是0:
log2_max_frame_num_minus4 = 0
pic_order_cnt_type = 0
log2_max_pic_order_cnt_lsb_minus4 = 0
4.现在第3步用掉了4个1,要从0001 01100001开始了
这里继续用Golomb解析max_num_ref_frames,很明显前缀有3个0。所以codenum=2^3-1+read_bits(011)=10
5.后面不再细述,结果如下
gaps_in_frame_num_value_allowed_flag = 0 // 0
pic_width_in_mbs_minus1 = 10 // Exp-Golomb解0001 011
pic_height_in_map_units_minus1 = 8 // Exp-Golomb解00010 01
frame_mbs_only_flag = 1 // 1
direct_8x8_inference_flag = 0 // 0
frame_cropping_flag = 0 // 0
vui_parameters_present_flag = 0 // 0
还剩下10两个位的数据没有用到,之前的这么多数据(除了NALU头之外的)都是seq_parameter_set_data,而根据Spec我们知道还有结尾补齐位
seq_parameter_set_rbsp( ) {
seq_parameter_set_data( ) // 数据
rbsp_trailing_bits( ) // 按字节补齐
}
补齐规则参见7.3.2.11 RBSP trailing bits syntax,实际就是按照字节对齐来补齐,所以这就是10这两位数据的由来。
回头看起来,这就是SPS的数据,也就是第一个NALU,同前面从Headers Info拷贝出来的SPS也是完全吻合的,所以这里我们就算是把Spec和实际的用法/码流对照起来了。另外值得说一下的就是从Headers Info拷贝出来的数据当中”na”就是未定义的,也就是if条件没有覆盖的情况。
二、PPS
这里PPS结构可以参考H264系列十六 句法元素PPS
去除掉NALU开始标记后的68 C8 A1 43 88
,转换为二进制是01101000 11001000 10100001 01000011 10001000
。
1.先看NALU头
forbidden_zero_bit = 0 // 0
nal_ref_idc = 3 // 11
nal_unit_type = 8 // 01000
也就对应于#define NALU_TYPE_PPS 8
,就可以知道此处的RBSP是PPS。
2.解码结果
pic_parameter_set_id = 0 // Exp-Golomb解1
seq_parameter_set_id = 0 // Exp-Golomb解1
entropy_coding_mode_flag = 0 // 0
bottom_field_pic_order_in_frame_present_flag = 0 // 0
num_slice_groups_minus1 = 0 // Exp-Golomb解1
num_ref_idx_l0_default_active_minus1 = 9 // Exp-Golomb解000 1010
num_ref_idx_l1_default_active_minus1 = 9 // Exp-Golomb解0001 010
weighted_pred_flag = 0 // 0
weighted_bipred_idc = 0 // 00
pic_init_qp_minus26 = 0 // Exp-Golomb解1
pic_init_qs_minus26 = 0 // Exp-Golomb解1
chroma_qp_index_offset = 0 // Exp-Golomb解1
deblocking_filter_control_present_flag = 0 // 0
constrained_intra_pred_flag = 0 // 0
redundant_pic_cnt_present_flag = 0 // 0
还剩下1000这四位,这就是按字节补齐的数据。
三、Slice
这里Slice结构可以参考H264系列十七 Slice(条带或片层)
去除掉NALU开始标记后的65 88 84 02
,转换为二进制是01100101 10001000 10000100 00000010
。
1.先看NALU头
forbidden_zero_bit = 0 // 0
nal_ref_idc = 3 // 11
nal_unit_type = 5 // 00101
也就对应于#define NALU_TYPE_IDR 5
,可以知道这个是IDR帧
first_mb_in_slice = 0 // Exp-Golomb解1
slice_type = 7 // Exp-Golomb解0001000
slice_type =7说明是个I帧
pic_parameter_set_id = 0 // Exp-Golomb解1
frame_num = 0 // u(v)根据占用的位数
//(log2_max_frame_num_minus4 + 4)解出值 // 0000
对于frame_num这个属性要特别说下,它的Descriptor是u(v),那么我们查看u(v)得知
u(n): unsigned integer using n bits. When n is “v” in the syntax table, the number of bits varies in a manner dependent on the value of other syntax elements.
也就是说这个属性占用的位数是取决于其它属性的,那就再搜索下frame_num得到
frame_num is used as an identifier for pictures and shall be represented by log2_max_frame_num_minus4 + 4 bits in the bitstream.
于是我们就大概清楚了,frame_num占用的位数跟log2_max_frame_num_minus4相关,之前在SPS当中我们知道log2_max_frame_num_minus4 = 0,所以这里frame_num占用4位,也就是0000,解析出来也就是0,另外也需要知道frame_num有很多限制,比如在IDR当中必须为0,具体参见7.4.3 Slice header semantics。这里要指出的是,这是一份完整优秀的Spec,基本上已经涵盖了我们需要的所有东西,只是需要我们去找,去分析(尽管这个过程可能很麻烦,有时让人摸不着头脑,但是需要相信我们需要的答案就在里面)。
idr_pic_id = 0 // Exp-Golomb解1
pic_order_cnt_lsb = 0 // u(v)根据占用的位数
//(log2_max_pic_order_cnt_lsb_minus4 + 4)解出值 // 0000
剩下的000010。现在要进入ref_pic_list_modification( )这个function了,但是里面所有if判断条件不符合
然后进入
dec_ref_pic_marking( )
no_output_of_prior_pics_flag = 0 // 0
long_term_reference_flag = 0 // 0
现在只剩下0010这四位了,我们继续补充3个字节(63 61 7C)进来01100011 01100001 01111100
。于是我们继续做slice_qp_delta的解码,注意这里它的Descriptor是se(v),所以要先对进Exp-Golomb解码,然后进行mapping得出值。
0010 01100011 01100001 01111100
// 这里两个0,求出Exp-Golomb编码值为00100 // 长度5,后缀为0可以被解析成2 实际可以通过Exp-Golomb(2^2 – 1 + 0)算出值为3 然后代入(-1)^(k + 1) * Ceil(k divide 2)求出值为2。详细可以参见9.1.1 Mapping process for signed Exp-Golomb codes。
slice_qp_delta = 2 // 00100 // se(v)
到这里Slice Header就解析完成了。
暂时就到这里,需要说明的是,我们只写出了前三个NALU部分解析方法(第一个Slice,也就是IDR,我们只写出了Header部分,还有数据部分我们留到后面来分析),还剩两个Slice我们留着有必要的时候来分析。