lavfi: Add blurdetect filter

Changelog

@@ -13,6 +13,7 @@ version 5.1:
 - pixelize video filter
 - colormap video filter
 - colorchart video source filter
+- blurdetect filter
 
 
 version 5.0:

doc/filters.texi

@@ -7997,6 +7997,58 @@ tblend=all_mode=grainextract
 @subsection Commands
 This filter supports same @ref{commands} as options.
 
+@anchor{blurdetect}
+@section blurdetect
+
+Determines blurriness of frames without altering the input frames.
+
+Based on Marziliano, Pina, et al. "A no-reference perceptual blur metric."
+Allows for a block-based abbreviation.
+
+The filter accepts the following options:
+
+@table @option
+@item low
+@item high
+Set low and high threshold values used by the Canny thresholding
+algorithm.
+
+The high threshold selects the "strong" edge pixels, which are then
+connected through 8-connectivity with the "weak" edge pixels selected
+by the low threshold.
+
+@var{low} and @var{high} threshold values must be chosen in the range
+[0,1], and @var{low} should be lesser or equal to @var{high}.
+
+Default value for @var{low} is @code{20/255}, and default value for @var{high}
+is @code{50/255}.
+
+@item radius
+Define the radius to search around an edge pixel for local maxima.
+
+@item block_pct
+Determine blurriness only for the most significant blocks, given in percentage.
+
+@item block_width
+Determine blurriness for blocks of width @var{block_width}. If set to any value smaller 1, no blocks are used and the whole image is processed as one no matter of @var{block_height}.
+
+@item block_height
+Determine blurriness for blocks of height @var{block_height}. If set to any value smaller 1, no blocks are used and the whole image is processed as one no matter of @var{block_width}.
+
+@item planes
+Set planes to filter. Default is first only.
+@end table
+
+@subsection Examples
+
+@itemize
+@item
+Determine blur for 80% of most significant 32x32 blocks:
+@example
+blurdetect=block_width=32:block_height=32:block_pct=80
+@end example
+@end itemize
+
 @section bm3d
 
 Denoise frames using Block-Matching 3D algorithm.

libavfilter/Makefile

@@ -195,6 +195,7 @@ OBJS-$(CONFIG_BLACKDETECT_FILTER)            += vf_blackdetect.o
 OBJS-$(CONFIG_BLACKFRAME_FILTER)             += vf_blackframe.o
 OBJS-$(CONFIG_BLEND_FILTER)                  += vf_blend.o framesync.o
 OBJS-$(CONFIG_BLEND_VULKAN_FILTER)           += vf_blend_vulkan.o framesync.o vulkan.o vulkan_filter.o
+OBJS-$(CONFIG_BLURDETECT_FILTER)             += vf_blurdetect.o edge_common.o
 OBJS-$(CONFIG_BM3D_FILTER)                   += vf_bm3d.o framesync.o
 OBJS-$(CONFIG_BOXBLUR_FILTER)                += vf_boxblur.o boxblur.o
 OBJS-$(CONFIG_BOXBLUR_OPENCL_FILTER)         += vf_avgblur_opencl.o opencl.o \

libavfilter/allfilters.c

@@ -183,6 +183,7 @@ extern const AVFilter ff_vf_blackdetect;
 extern const AVFilter ff_vf_blackframe;
 extern const AVFilter ff_vf_blend;
 extern const AVFilter ff_vf_blend_vulkan;
+extern const AVFilter ff_vf_blurdetect;
 extern const AVFilter ff_vf_bm3d;
 extern const AVFilter ff_vf_boxblur;
 extern const AVFilter ff_vf_boxblur_opencl;

libavfilter/version.h

@@ -31,7 +31,7 @@
 
 #include "version_major.h"
 
-#define LIBAVFILTER_VERSION_MINOR  36
+#define LIBAVFILTER_VERSION_MINOR  37
 #define LIBAVFILTER_VERSION_MICRO 100
 
 

libavfilter/vf_blurdetect.c

@@ -0,0 +1,394 @@
+/*
+ * Copyright (c) 2021 Thilo Borgmann <thilo.borgmann _at_ mail.de>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * No-reference blurdetect filter
+ *
+ * Implementing:
+ * Marziliano, Pina, et al. "A no-reference perceptual blur metric." Proceedings.
+ * International conference on image processing. Vol. 3. IEEE, 2002.
+ * https://infoscience.epfl.ch/record/111802/files/14%20A%20no-reference%20perceptual%20blur%20metric.pdf
+ *
+ * @author Thilo Borgmann <thilo.borgmann _at_ mail.de>
+ */
+
+#include "libavutil/avassert.h"
+#include "libavutil/imgutils.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixelutils.h"
+#include "libavutil/motion_vector.h"
+#include "libavutil/qsort.h"
+#include "avfilter.h"
+#include "formats.h"
+#include "internal.h"
+#include "video.h"
+#include "edge_common.h"
+
+static int comp(const float *a,const float *b)
+{
+    return FFDIFFSIGN(*a, *b);
+}
+
+typedef struct BLRContext {
+    const AVClass *class;
+
+    int hsub, vsub;
+    int nb_planes;
+
+    float   low, high;
+    uint8_t low_u8, high_u8;
+    int     radius;        // radius during local maxima detection
+    int     block_pct;     // percentage of "sharpest" blocks in the image to use for bluriness calculation
+    int     block_width;   // width for block abbreviation
+    int     block_height;  // height for block abbreviation
+    int     planes;        // number of planes to filter
+
+    double   blur_total;
+    uint64_t nb_frames;
+
+    float    *blks;
+    uint8_t  *filterbuf;
+    uint8_t  *tmpbuf;
+    uint16_t *gradients;
+    char     *directions;
+} BLRContext;
+
+#define OFFSET(x) offsetof(BLRContext, x)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+static const AVOption blurdetect_options[] = {
+    { "high",          "set high threshold", OFFSET(high), AV_OPT_TYPE_FLOAT, {.dbl=30/255.}, 0, 1, FLAGS },
+    { "low",           "set low threshold",  OFFSET(low),  AV_OPT_TYPE_FLOAT, {.dbl=15/255.}, 0, 1, FLAGS },
+    { "radius",        "search radius for maxima detection", OFFSET(radius), AV_OPT_TYPE_INT, {.i64=50}, 1, 100, FLAGS },
+    { "block_pct",     "block pooling threshold when calculating blurriness", OFFSET(block_pct), AV_OPT_TYPE_INT, {.i64=80}, 1, 100, FLAGS },
+    { "block_width",   "block size for block-based abbreviation of blurriness", OFFSET(block_width), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX, FLAGS },
+    { "block_height",  "block size for block-based abbreviation of blurriness", OFFSET(block_height), AV_OPT_TYPE_INT, {.i64=-1}, -1, INT_MAX, FLAGS },
+    { "planes",        "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=1}, 0, 15, FLAGS },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(blurdetect);
+
+static av_cold int blurdetect_init(AVFilterContext *ctx)
+{
+    BLRContext *s = ctx->priv;
+
+    s->low_u8  = s->low  * 255. + .5;
+    s->high_u8 = s->high * 255. + .5;
+
+    return 0;
+}
+
+static int blurdetect_config_input(AVFilterLink *inlink)
+{
+    AVFilterContext *ctx = inlink->dst;
+    BLRContext      *s   = ctx->priv;
+    const int bufsize    = inlink->w * inlink->h;
+    const AVPixFmtDescriptor *pix_desc;
+
+    pix_desc = av_pix_fmt_desc_get(inlink->format);
+    s->hsub = pix_desc->log2_chroma_w;
+    s->vsub = pix_desc->log2_chroma_h;
+    s->nb_planes = av_pix_fmt_count_planes(inlink->format);
+
+    if (s->block_width  < 1 || s->block_height < 1) {
+        s->block_width  = inlink->w;
+        s->block_height = inlink->h;
+    }
+
+    s->tmpbuf     = av_malloc(bufsize);
+    s->filterbuf  = av_malloc(bufsize);
+    s->gradients  = av_calloc(bufsize, sizeof(*s->gradients));
+    s->directions = av_malloc(bufsize);
+    s->blks       = av_calloc((inlink->w / s->block_width) * (inlink->h / s->block_height),
+                              sizeof(*s->blks));
+
+    if (!s->tmpbuf || !s->filterbuf || !s->gradients || !s->directions || !s->blks)
+        return AVERROR(ENOMEM);
+
+    return 0;
+}
+
+// edge width is defined as the distance between surrounding maxima of the edge pixel
+static float edge_width(BLRContext *blr, int i, int j, int8_t dir, int w, int h,
+                        int edge, const uint8_t *src, int src_linesize)
+{
+    float width = 0;
+    int dX, dY;
+    int sign;
+    int tmp;
+    int p1;
+    int p2;
+    int k, x, y;
+    int edge1;
+    int edge2;
+    float luma1 = 0.0; // average luma difference per edge pixel
+    float luma2 = 0.0;
+    int radius = blr->radius;
+
+    switch(dir) {
+    case DIRECTION_HORIZONTAL: dX = 1; dY =  0; break;
+    case DIRECTION_VERTICAL:   dX = 0; dY =  1; break;
+    case DIRECTION_45UP:       dX = 1; dY = -1; break;
+    case DIRECTION_45DOWN:     dX = 1; dY =  1; break;
+    }
+
+    // determines if search in direction dX/dY is looking for a maximum or minimum
+    sign = src[j * src_linesize + i] > src[(j - dY) * src_linesize + i - dX] ? 1 : -1;
+
+    // search in -(dX/dY) direction
+    for (k = 0; k < radius; k++) {
+        x = i - k*dX;
+        y = j - k*dY;
+        p1 = y * src_linesize + x;
+        x -= dX;
+        y -= dY;
+        p2 = y * src_linesize + x;
+        if (x < 0 || x >= w || y < 0 || y >= h)
+            return 0;
+
+        tmp = (src[p1] - src[p2]) * sign;
+
+        if (tmp <= 0) // local maximum found
+            break;
+
+        luma1 += tmp;
+    }
+    if (k > 0) luma1 /= k;
+    edge1 = k;
+    width += k;
+
+    // search in +(dX/dY) direction
+    for (k = 0; k < radius; k++) {
+        x = i + k * dX;
+        y = j + k * dY;
+        p1 = y * src_linesize + x;
+        x += dX;
+        y += dY;
+        p2 = y * src_linesize + x;
+        if (x < 0 || x >= w || y < 0 || y >= h)
+            return 0;
+
+        tmp = (src[p1] - src[p2]) * sign;
+
+        if (tmp >= 0) // local maximum found
+            break;
+
+        luma2 -= tmp;
+    }
+    if (k > 0) luma2 /= k;
+    edge2 = k;
+    width += k;
+
+    // for 45 degree directions approximate edge width in pixel units: 0.7 ~= sqrt(2)/2
+    if (dir == DIRECTION_45UP || dir == DIRECTION_45DOWN)
+        width *= 0.7;
+
+    return width;
+}
+
+static float calculate_blur(BLRContext *s, int w, int h, int hsub, int vsub,
+                            uint8_t* dir, int dir_linesize,
+                            uint8_t* dst, int dst_linesize,
+                            uint8_t* src, int src_linesize)
+{
+    float total_width = 0.0;
+    int block_count;
+    double block_total_width;
+
+    int i, j;
+    int blkcnt = 0;
+
+    float *blks = s->blks;
+    float block_pool_threshold = s->block_pct / 100.0;
+
+    int block_width  = AV_CEIL_RSHIFT(s->block_width,  hsub);
+    int block_height = AV_CEIL_RSHIFT(s->block_height, vsub);
+    int brows = h / block_height;
+    int bcols = w / block_width;
+
+    for (int blkj = 0; blkj < brows; blkj++) {
+        for (int blki = 0; blki < bcols; blki++) {
+            block_total_width = 0.0;
+            block_count = 0;
+            for (int inj = 0; inj < block_height; inj++) {
+                for (int ini = 0; ini < block_width; ini++) {
+                    i = blki * block_width + ini;
+                    j = blkj * block_height + inj;
+
+                    if (dst[j * dst_linesize + i] > 0) {
+                        float width = edge_width(s, i, j, dir[j*dir_linesize+i],
+                                                 w, h, dst[j*dst_linesize+i],
+                                                 src, src_linesize);
+                        if (width > 0.001) { // throw away zeros
+                            block_count++;
+                            block_total_width += width;
+                        }
+                    }
+                }
+            }
+            // if not enough edge pixels in a block, consider it smooth
+            if (block_total_width >= 2) {
+                blks[blkcnt] = block_total_width / block_count;
+                blkcnt++;
+            }
+        }
+    }
+
+    // simple block pooling by sorting and keeping the sharper blocks
+    AV_QSORT(blks, blkcnt, float, comp);
+    blkcnt = ceil(blkcnt * block_pool_threshold);
+    for (int i = 0; i < blkcnt; i++) {
+        total_width += blks[i];
+    }
+
+    return  total_width / blkcnt;
+}
+
+static void set_meta(AVDictionary **metadata, const char *key, float d)
+{
+    char value[128];
+    snprintf(value, sizeof(value), "%f", d);
+    av_dict_set(metadata, key, value, 0);
+}
+
+static int blurdetect_filter_frame(AVFilterLink *inlink, AVFrame *in)
+{
+    AVFilterContext *ctx  = inlink->dst;
+    BLRContext *s         = ctx->priv;
+    AVFilterLink *outlink = ctx->outputs[0];
+
+    const int inw = inlink->w;
+    const int inh = inlink->h;
+
+    uint8_t *tmpbuf     = s->tmpbuf;
+    uint8_t *filterbuf  = s->filterbuf;
+    uint16_t *gradients = s->gradients;
+    int8_t *directions  = s->directions;
+
+    float blur = 0.0f;
+    int nplanes = 0;
+    AVDictionary **metadata;
+    metadata = &in->metadata;
+
+    for (int plane = 0; plane < s->nb_planes; plane++) {
+        int hsub = plane == 1 || plane == 2 ? s->hsub : 0;
+        int vsub = plane == 1 || plane == 2 ? s->vsub : 0;
+        int w = AV_CEIL_RSHIFT(inw, hsub);
+        int h = AV_CEIL_RSHIFT(inh, vsub);
+
+        if (!((1 << plane) & s->planes))
+            continue;
+
+        nplanes++;
+
+        // gaussian filter to reduce noise
+        ff_gaussian_blur(w, h,
+                         filterbuf,  w,
+                         in->data[plane], in->linesize[plane]);
+
+        // compute the 16-bits gradients and directions for the next step
+        ff_sobel(w, h, gradients, w, directions, w, filterbuf, w);
+
+        // non_maximum_suppression() will actually keep & clip what's necessary and
+        // ignore the rest, so we need a clean output buffer
+        memset(tmpbuf, 0, inw * inh);
+        ff_non_maximum_suppression(w, h, tmpbuf, w, directions, w, gradients, w);
+
+
+        // keep high values, or low values surrounded by high values
+        ff_double_threshold(s->low_u8, s->high_u8, w, h,
+                            tmpbuf, w, tmpbuf, w);
+
+        blur += calculate_blur(s, w, h, hsub, vsub, directions, w,
+                              tmpbuf, w, filterbuf, w);
+    }
+
+    if (nplanes)
+        blur /= nplanes;
+
+    s->blur_total += blur;
+
+    // write stats
+    av_log(ctx, AV_LOG_VERBOSE, "blur: %.7f\n", blur);
+
+    set_meta(metadata, "lavfi.blur", blur);
+
+    s->nb_frames = inlink->frame_count_in;
+
+    return ff_filter_frame(outlink, in);
+}
+
+static av_cold void blurdetect_uninit(AVFilterContext *ctx)
+{
+    BLRContext *s = ctx->priv;
+
+    if (s->nb_frames > 0) {
+        av_log(ctx, AV_LOG_INFO, "blur mean: %.7f\n",
+               s->blur_total / s->nb_frames);
+    }
+
+    av_freep(&s->tmpbuf);
+    av_freep(&s->filterbuf);
+    av_freep(&s->gradients);
+    av_freep(&s->directions);
+    av_freep(&s->blks);
+}
+
+static const enum AVPixelFormat pix_fmts[] = {
+    AV_PIX_FMT_GRAY8,
+    AV_PIX_FMT_GBRP,     AV_PIX_FMT_GBRAP,
+    AV_PIX_FMT_YUV422P,  AV_PIX_FMT_YUV420P,
+    AV_PIX_FMT_YUV444P,  AV_PIX_FMT_YUV440P,
+    AV_PIX_FMT_YUV411P,  AV_PIX_FMT_YUV410P,
+    AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUVJ420P,
+    AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ444P,
+    AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA420P,
+    AV_PIX_FMT_NONE
+};
+
+static const AVFilterPad blurdetect_inputs[] = {
+    {
+        .name         = "default",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .config_props = blurdetect_config_input,
+        .filter_frame = blurdetect_filter_frame,
+    },
+};
+
+static const AVFilterPad blurdetect_outputs[] = {
+    {
+        .name = "default",
+        .type = AVMEDIA_TYPE_VIDEO,
+    },
+};
+
+const AVFilter ff_vf_blurdetect = {
+    .name          = "blurdetect",
+    .description   = NULL_IF_CONFIG_SMALL("Blurdetect filter."),
+    .priv_size     = sizeof(BLRContext),
+    .init          = blurdetect_init,
+    .uninit        = blurdetect_uninit,
+    FILTER_PIXFMTS_ARRAY(pix_fmts),
+    FILTER_INPUTS(blurdetect_inputs),
+    FILTER_OUTPUTS(blurdetect_outputs),
+    .priv_class    = &blurdetect_class,
+    .flags         = AVFILTER_FLAG_METADATA_ONLY,
+};

tests/fate/filter-video.mak

@@ -867,6 +867,9 @@ fate-filter-meta-4560-rotate0: CMD = framecrc -auto_conversion_filters -flags +b
 
 REFCMP_DEPS = FFMPEG LAVFI_INDEV TESTSRC2_FILTER AVGBLUR_FILTER METADATA_FILTER
 
+FATE_FILTER-$(call ALLYES, $(REFCMP_DEPS) BLURDETECT_FILTER) += fate-filter-refcmp-blurdetect-yuv
+fate-filter-refcmp-blurdetect-yuv: CMD = cmp_metadata blurdetect yuv420p 0.015
+
 FATE_FILTER-$(call ALLYES, $(REFCMP_DEPS) PSNR_FILTER) += fate-filter-refcmp-psnr-rgb
 fate-filter-refcmp-psnr-rgb: CMD = refcmp_metadata psnr rgb24 0.002
 

tests/ref/fate/filter-refcmp-blurdetect-yuv

@@ -0,0 +1,10 @@
+frame:0    pts:0       pts_time:0
+lavfi.blur=4.499666
+frame:1    pts:1       pts_time:1
+lavfi.blur=4.677492
+frame:2    pts:2       pts_time:2
+lavfi.blur=4.735711
+frame:3    pts:3       pts_time:3
+lavfi.blur=4.532343
+frame:4    pts:4       pts_time:4
+lavfi.blur=4.532660