[FFmpeg-devel] [PATCH] lavf/vf_ocr: add subregion support

[Lingjiang Fang]

---
 doc/filters.texi     |  7 +++++++
 libavfilter/vf_ocr.c | 30 +++++++++++++++++++++++++++++-
 2 files changed, 36 insertions(+), 1 deletion(-)

diff --git a/doc/filters.texi b/doc/filters.texi
index da8f7d7726..9c650a2a5a 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -15451,6 +15451,13 @@ Set character whitelist.
 
 @item blacklist
 Set character blacklist.
+
+ at item x, y
+Set top point position of subregion, not support expression now
+
+ at item w, h
+Set Width and height of subregion
+
 @end table
 
 The filter exports recognized text as the frame metadata @code{lavfi.ocr.text}.
diff --git a/libavfilter/vf_ocr.c b/libavfilter/vf_ocr.c
index 6de474025a..7beb101679 100644
--- a/libavfilter/vf_ocr.c
+++ b/libavfilter/vf_ocr.c
@@ -33,6 +33,8 @@ typedef struct OCRContext {
     char *language;
     char *whitelist;
     char *blacklist;
+    int x, y;
+    int w, h;
 
     TessBaseAPI *tess;
 } OCRContext;
@@ -45,6 +47,10 @@ static const AVOption ocr_options[] = {
     { "language",  "set language",            OFFSET(language),  AV_OPT_TYPE_STRING, {.str="eng"}, 0, 0, FLAGS },
     { "whitelist", "set character whitelist", OFFSET(whitelist), AV_OPT_TYPE_STRING, {.str="0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ.:;,-+_!?\"'[]{}()<>|/\\=*&%$#@!~ "}, 0, 0, FLAGS },
     { "blacklist", "set character blacklist", OFFSET(blacklist), AV_OPT_TYPE_STRING, {.str=""},    0, 0, FLAGS },
+    { "x",         "top x of sub region",     OFFSET(x),         AV_OPT_TYPE_INT,    {.i64=0},     0, INT_MAX, FLAGS },
+    { "y",         "top y of sub region",     OFFSET(y),         AV_OPT_TYPE_INT,    {.i64=0},     0, INT_MAX, FLAGS },
+    { "w",         "width of sub region",     OFFSET(w),         AV_OPT_TYPE_INT,    {.i64=0},     0, INT_MAX, FLAGS },
+    { "h",         "height of sub region",    OFFSET(h),         AV_OPT_TYPE_INT,    {.i64=0},     0, INT_MAX, FLAGS },
     { NULL }
 };
 
@@ -73,6 +79,19 @@ static av_cold int init(AVFilterContext *ctx)
     return 0;
 }
 
+static int config_input(AVFilterLink *inlink)
+{
+    OCRContext  *s = inlink->dst->priv;
+
+    // may call many times, we don't check w/h here
+    if (s->x < 0 || s->y < 0) {
+        s->x = 0;
+        s->y = 0;
+    }
+
+    return 0;
+}
+
 static int query_formats(AVFilterContext *ctx)
 {
     static const enum AVPixelFormat pix_fmts[] = {
@@ -101,9 +120,17 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
     OCRContext *s = ctx->priv;
     char *result;
     int *confs;
+    int w = s->w;
+    int h = s->h;
+
+    if (w <= 0 || h <= 0) {
+        w = in->width;
+        h = in->height;
+    }
 
+    av_log(s, AV_LOG_ERROR, "x=%d, y=%d, w=%d, h=%d\n", s->x, s->y, w, h);
     result = TessBaseAPIRect(s->tess, in->data[0], 1,
-                             in->linesize[0], 0, 0, in->width, in->height);
+                             in->linesize[0], s->x, s->y, w, h);
     confs = TessBaseAPIAllWordConfidences(s->tess);
     av_dict_set(metadata, "lavfi.ocr.text", result, 0);
     for (int i = 0; confs[i] != -1; i++) {
@@ -134,6 +161,7 @@ static const AVFilterPad ocr_inputs[] = {
         .name         = "default",
         .type         = AVMEDIA_TYPE_VIDEO,
         .filter_frame = filter_frame,
+        .config_props = config_input,
     },
     { NULL }
 };
-- 
2.29.2