opencv 之 icvCreateHidHaarClassifierCascade 分类器信息初始化函数部分详细代码注释。

请看注释。这个函数，是人脸识别主函数，里面出现过的函数之一，作用是初始化分类器的数据，就是一个xml文件的数据初始化。

---

static CvHidHaarClassifierCascade* icvCreateHidHaarClassifierCascade( CvHaarClassifierCascade* cascade )
{
    CvRect* ipp_features = 0;//定义一个矩形框指针
    float *ipp_weights = 0, *ipp_thresholds = 0, *ipp_val1 = 0, *ipp_val2 = 0;//单精度浮点数指针4个
    int* ipp_counts = 0;//整形指针1个

    CvHidHaarClassifierCascade* out = 0;//最终返回的值

    int i, j, k, l;//for循环的控制变量
    int datasize;//数据大小
    int total_classifiers = 0;//总的分类器数目
    int total_nodes = 0;
    char errorstr[1000];//错误信息数组
    CvHidHaarClassifier* haar_classifier_ptr;//级联分类器指针
    CvHidHaarTreeNode* haar_node_ptr;
    CvSize orig_window_size;//提取窗口的大小
    int has_tilted_features = 0;
    int max_count = 0;

    if( !CV_IS_HAAR_CLASSIFIER(cascade) )//判断传进来的分类器文件是否真正确
        CV_Error( !cascade ? CV_StsNullPtr : CV_StsBadArg, "Invalid classifier pointer" );

    if( cascade->hid_cascade )//判断改分类器xml文件是否已经被初始化了
        CV_Error( CV_StsError, "hid_cascade has been already created" );

    if( !cascade->stage_classifier )//如果没有阶级分类器，报错
        CV_Error( CV_StsNullPtr, "" );

    if( cascade->count <= 0 )//如果分类器的阶级数<=0，报错
        CV_Error( CV_StsOutOfRange, "Negative number of cascade stages" );

    orig_window_size = cascade->orig_window_size;//获取识别窗口的大小

    /* check input structure correctness and calculate total memory size needed for
       internal representation of the classifier cascade */

    for( i = 0; i < cascade->count; i++ )//对xml文件里面的每阶段的stage进行循环提取相关数据
    {
        CvHaarStageClassifier* stage_classifier = cascade->stage_classifier + i;
        //获取每次进入循环的后阶段的子分类器，以haarcascade_upperbody.xml 为例子，count是30，stage_classifier的count是20

        if( !stage_classifier->classifier ||//判断阶段分类器、子分类器及其stage 层数 是否合法
            stage_classifier->count <= 0 )
        {
            sprintf( errorstr, "header of the stage classifier #%d is invalid "
                     "(has null pointers or non-positive classfier count)", i );
            CV_Error( CV_StsError, errorstr );
        }

        max_count = MAX( max_count, stage_classifier->count );//获取子分类器stage的数目，以haarcascade_upperbody.xml为例，是20
        total_classifiers += stage_classifier->count;//统计出总的子分类器的stage数目，即tree，再统计

        for( j = 0; j < stage_classifier->count; j++ )
        //这个for循环主要是进入到子分类器tree里面的数据提取并且对其正确性的判断，
        //循环条件为字stage数目，以haarcascade_upperbody.xml为例，为20
        {
            CvHaarClassifier* classifier = stage_classifier->classifier + j;//同上，找到此时循环的tree

            total_nodes += classifier->count;//计算出此时循环的tree子分类器的root node 数目，再统计。以haarcascade_upperbody.xml为例，每个tree的node是1
            for( l = 0; l < classifier->count; l++ )
                //这个是关键循环，主数据的获取
                //以haarcascade_upperbody.xml为例，此时classifier->count=1，循环一次，进入里面获取关键数据
            {
                for( k = 0; k < CV_HAAR_FEATURE_MAX; k++ )//CV_HAAR_FEATURE_MAX = 3，循环三次，feature的最大数目，以haarcascade_upperbody.xml为例，只有1个
                {
                    if( classifier->haar_feature[l].rect[k].r.width )
                        //逐层递归，先找feature，再找它里面的rect标签里面的矩阵row，行的宽度
                        //以haarcascade_upperbody.xml为例，是2
                    {
                        CvRect r = classifier->haar_feature[l].rect[k].r;//把此时row矩阵框赋给r
                        int tilted = classifier->haar_feature[l].tilted;//获取xml标签tited的值
                        has_tilted_features |= tilted != 0;//|是位运算，例如0|1=1,这行的作用是判断has和tilted那个是1，还不知道其意义何在
                        if( r.width < 0 || r.height < 0 || r.y < 0 ||
                            r.x + r.width > orig_window_size.width
                            ||
                            (!tilted &&
                            (r.x < 0 || r.y + r.height > orig_window_size.height))
                            ||
                            (tilted && (r.x - r.height < 0 ||
                            r.y + r.width + r.height > orig_window_size.height)))
                            //这个if语句是对feature里面的数据矩形的各方面判断，包括矩形的宽、高、等
                            //矩形# %d的分类器# %d”“级分类器# %d是不是在里面”“参考（原创）级联窗口”
                        {
                            sprintf( errorstr, "rectangle #%d of the classifier #%d of "
                                     "the stage classifier #%d is not inside "
                                     "the reference (original) cascade window", k, j, i );
                            CV_Error( CV_StsNullPtr, errorstr );
                        }
                    }
                }
            }
        }
    }
    //上面数据的判断结束后，到这里

    datasize = sizeof(CvHidHaarClassifierCascade) +//获取整个分类器，xml文件的数据大小
               sizeof(CvHidHaarStageClassifier)*cascade->count +
               sizeof(CvHidHaarClassifier) * total_classifiers +
               sizeof(CvHidHaarTreeNode) * total_nodes +
               sizeof(void*)*(total_nodes + total_classifiers);

    out = (CvHidHaarClassifierCascade*)cvAlloc( datasize );//给最终返回的变量分配内存大小
    memset( out, 0, sizeof(*out) );//对变量初始化，全部填充0

    //下面是逐个赋值,初始化头部
    /* init header */
    out->count = cascade->count;//新分类器out的stage数目
    out->stage_classifier = (CvHidHaarStageClassifier*)(out + 1);//子分类器tree的数目
    haar_classifier_ptr = (CvHidHaarClassifier*)(out->stage_classifier + cascade->count);//tree指针
    haar_node_ptr = (CvHidHaarTreeNode*)(haar_classifier_ptr + total_classifiers);//tree里面node的指针

    out->isStumpBased = 1;//布尔类型，true
    out->has_tilted_features = has_tilted_features;
    out->is_tree = 0;

    /* initialize internal representation */
    for( i = 0; i < cascade->count; i++ )
    {
        CvHaarStageClassifier* stage_classifier = cascade->stage_classifier + i;
        CvHidHaarStageClassifier* hid_stage_classifier = out->stage_classifier + i;

        hid_stage_classifier->count = stage_classifier->count;
        hid_stage_classifier->threshold = stage_classifier->threshold - icv_stage_threshold_bias;
        hid_stage_classifier->classifier = haar_classifier_ptr;
        hid_stage_classifier->two_rects = 1;
        haar_classifier_ptr += stage_classifier->count;

        hid_stage_classifier->parent = (stage_classifier->parent == -1)
            ? NULL : out->stage_classifier + stage_classifier->parent;
        hid_stage_classifier->next = (stage_classifier->next == -1)
            ? NULL : out->stage_classifier + stage_classifier->next;
        hid_stage_classifier->child = (stage_classifier->child == -1)
            ? NULL : out->stage_classifier + stage_classifier->child;

        out->is_tree |= hid_stage_classifier->next != NULL;

        for( j = 0; j < stage_classifier->count; j++ )
        {
            CvHaarClassifier* classifier = stage_classifier->classifier + j;
            CvHidHaarClassifier* hid_classifier = hid_stage_classifier->classifier + j;
            int node_count = classifier->count;
            float* alpha_ptr = (float*)(haar_node_ptr + node_count);

            hid_classifier->count = node_count;
            hid_classifier->node = haar_node_ptr;
            hid_classifier->alpha = alpha_ptr;

            for( l = 0; l < node_count; l++ )
            {
                CvHidHaarTreeNode* node = hid_classifier->node + l;
                CvHaarFeature* feature = classifier->haar_feature + l;
                memset( node, -1, sizeof(*node) );
                node->threshold = classifier->threshold[l];
                node->left = classifier->left[l];
                node->right = classifier->right[l];

                if( fabs(feature->rect[2].weight) < DBL_EPSILON ||
                    feature->rect[2].r.width == 0 ||
                    feature->rect[2].r.height == 0 )
                    memset( &(node->feature.rect[2]), 0, sizeof(node->feature.rect[2]) );
                else
                    hid_stage_classifier->two_rects = 0;
            }

            memcpy( alpha_ptr, classifier->alpha, (node_count+1)*sizeof(alpha_ptr[0]));
            haar_node_ptr =
                (CvHidHaarTreeNode*)cvAlignPtr(alpha_ptr+node_count+1, sizeof(void*));

            out->isStumpBased &= node_count == 1;
        }
    }
/*
#ifdef HAVE_IPP
    int can_use_ipp = !out->has_tilted_features && !out->is_tree && out->isStumpBased;

    if( can_use_ipp )
    {
        int ipp_datasize = cascade->count*sizeof(out->ipp_stages[0]);
        float ipp_weight_scale=(float)(1./((orig_window_size.width-icv_object_win_border*2)*
            (orig_window_size.height-icv_object_win_border*2)));

        out->ipp_stages = (void**)cvAlloc( ipp_datasize );
        memset( out->ipp_stages, 0, ipp_datasize );

        ipp_features = (CvRect*)cvAlloc( max_count*3*sizeof(ipp_features[0]) );
        ipp_weights = (float*)cvAlloc( max_count*3*sizeof(ipp_weights[0]) );
        ipp_thresholds = (float*)cvAlloc( max_count*sizeof(ipp_thresholds[0]) );
        ipp_val1 = (float*)cvAlloc( max_count*sizeof(ipp_val1[0]) );
        ipp_val2 = (float*)cvAlloc( max_count*sizeof(ipp_val2[0]) );
        ipp_counts = (int*)cvAlloc( max_count*sizeof(ipp_counts[0]) );

        for( i = 0; i < cascade->count; i++ )
        {
            CvHaarStageClassifier* stage_classifier = cascade->stage_classifier + i;
            for( j = 0, k = 0; j < stage_classifier->count; j++ )
            {
                CvHaarClassifier* classifier = stage_classifier->classifier + j;
                int rect_count = 2 + (classifier->haar_feature->rect[2].r.width != 0);

                ipp_thresholds[j] = classifier->threshold[0];
                ipp_val1[j] = classifier->alpha[0];
                ipp_val2[j] = classifier->alpha[1];
                ipp_counts[j] = rect_count;

                for( l = 0; l < rect_count; l++, k++ )
                {
                    ipp_features[k] = classifier->haar_feature->rect[l].r;
                    //ipp_features[k].y = orig_window_size.height - ipp_features[k].y - ipp_features[k].height;
                    ipp_weights[k] = classifier->haar_feature->rect[l].weight*ipp_weight_scale;
                }
            }

            if( ippiHaarClassifierInitAlloc_32f( (IppiHaarClassifier_32f**)&out->ipp_stages[i],
                (const IppiRect*)ipp_features, ipp_weights, ipp_thresholds,
                ipp_val1, ipp_val2, ipp_counts, stage_classifier->count ) < 0 )
                break;
        }

        if( i < cascade->count )
        {
            for( j = 0; j < i; j++ )
                if( out->ipp_stages[i] )
                    ippiHaarClassifierFree_32f( (IppiHaarClassifier_32f*)out->ipp_stages[i] );
            cvFree( &out->ipp_stages );
        }
    }
#endif
*/
    cascade->hid_cascade = out;
    assert( (char*)haar_node_ptr - (char*)out <= datasize );

    cvFree( &ipp_features );
    cvFree( &ipp_weights );
    cvFree( &ipp_thresholds );
    cvFree( &ipp_val1 );
    cvFree( &ipp_val2 );
    cvFree( &ipp_counts );

    return out;
}