[FFmpeg-devel] [PATCH] Apple Video Encoder (rpza)
Vitor Sessak
vitor1001
Sun Jan 23 18:06:17 CET 2011
On 09/02/2007 09:15 PM, Ramiro Polla wrote:
> Todd Kirby wrote:
>> On 6/5/05, Mike Melanson<mike at multimedia.cx> wrote:
>>
>>> Todd Kirby wrote:
>>>
>>>> Yeah, that's what my target was. I needed something that could play on
>>>> even the oldest version of Quicktime Player. I used your excellent
>>>> description of the rpza format at...
>>>>
>>>> http://www.pcisys.net/~melanson/codecs/
>>>>
>>>> ...as a guide. I notice this document not there anymore. If you've
>>>> moved it, let me know and I'll change my link in the source.
>>>>
>>> The document is located @ multimedia.cx. I need to update all of those
>>> description links.
>>>
>>
>> Here's a patch to update the description links.
>>
>
> Well, this never got applied... Is there more to it now or is it still good?
Git-friendly patch attached so patchwork will catch it up.
-Vitor
>From ffmpeg.phpatgmail.com Sun Jan 23 18:05:03 2011
From: ffmpeg.phpatgmail.com (Todd Kirby)
Date: Sun, 23 Jan 2011 18:05:03 +0100
Subject: [PATCH] Apple Video Encoder (rpza)
Message-ID: <mailman.467.1295802395.1307.ffmpeg-devel at mplayerhq.hu>
---
libavcodec/Makefile | 3 +
libavcodec/allcodecs.c | 3 +
libavcodec/avcodec.h | 1 +
libavcodec/rpzaenc.c | 1169 ++++++++++++++++++++++++++++++++++++++++++++++++
libavformat/movenc.c | 1 +
5 files changed, 1177 insertions(+), 0 deletions(-)
create mode 100644 libavcodec/rpzaenc.c
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 9e38eef..d75d66a 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -136,6 +136,9 @@ endif
ifneq ($(CONFIG_SVQ1_DECODER)$(CONFIG_SVQ1_ENCODER),)
OBJS+= svq1.o
endif
+ifneq ($(CONFIG_RPZA_ENCODER),)
+ OBJS+= rpzaenc.o
+endif
ifeq ($(CONFIG_TRUEMOTION1_DECODER),yes)
OBJS+= truemotion1.o
endif
diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c
index a0a7d7f..5e0d3f9 100644
--- a/libavcodec/allcodecs.c
+++ b/libavcodec/allcodecs.c
@@ -189,6 +189,9 @@ void avcodec_register_all(void)
#ifdef CONFIG_LIBGSM
register_avcodec(&libgsm_encoder);
#endif //CONFIG_LIBGSM
+#ifdef CONFIG_RPZA_ENCODER
+ register_avcodec(&rpza_encoder);
+#endif //CONFIG_RPZA_ENCODER
#endif /* CONFIG_ENCODERS */
#ifdef CONFIG_RAWVIDEO_ENCODER
register_avcodec(&rawvideo_encoder);
diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
index 2abb391..60c0158 100644
--- a/libavcodec/avcodec.h
+++ b/libavcodec/avcodec.h
@@ -1943,6 +1943,7 @@ extern AVCodec sonic_encoder;
extern AVCodec sonic_ls_encoder;
extern AVCodec svq1_encoder;
extern AVCodec x264_encoder;
+extern AVCodec rpza_encoder;
extern AVCodec h263_decoder;
extern AVCodec h261_decoder;
diff --git a/libavcodec/rpzaenc.c b/libavcodec/rpzaenc.c
new file mode 100644
index 0000000..d8b5ab7
--- /dev/null
+++ b/libavcodec/rpzaenc.c
@@ -0,0 +1,1169 @@
+/*
+ * Quicktime RPZA Video Encoder.
+ * Copyright (C) 2005 the ffmpeg project
+ *
+ * This library 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 of the License, or (at your option) any later version.
+ *
+ * This library 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 this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+/**
+ * @file rpzaenc.c
+ * QT RPZA Video Encoder by Todd Kirby <doubleshot at pacbell.net> and David Adler
+ *
+ * For more information about the RPZA format, visit:
+ * http://www.pcisys.net/~melanson/codecs/
+ */
+
+#include "avcodec.h"
+#include "dsputil.h"
+#include "bitstream.h"
+#include "assert.h"
+
+#ifdef CONFIG_ENCODERS
+
+typedef struct RpzaContext {
+
+ AVCodecContext *avctx;
+ DSPContext dsp;
+
+ AVFrame current_frame; // buffer for current 24 bit source frame
+ AVFrame prev_frame; // buffer for previous 24 bit source frame
+ PutBitContext pb; // buffer for encoded frame data.
+
+ int frame_width; // width in pixels of source frame
+ int frame_height; // height in pixesl of source frame
+
+ unsigned char *buf;
+ int first_frame; // flag set to one when the first frame is being processed
+ // so that comparisons with previous frame data in not attempted
+
+#ifdef DEBUG_STATS
+ int one_count, one_blocks_count;
+ int four_count, sixteen_count;
+ int skip_count, skip_blocks_count;
+#endif
+} RpzaContext;
+
+
+typedef struct rgb {
+ uint8_t r;
+ uint8_t g;
+ uint8_t b;
+} rgb;
+
+#define PIXELSTRIDE 3
+
+#define MIN(a,b) ((a) < (b) ? (a) : (b))
+#define MAX(a,b) ((a) > (b) ? (a) : (b))
+
+#define SQR(x) ((x) * (x))
+
+/* 15 bit components */
+#define GET_CHAN(color, chan) ((color) >> ((chan) * 5) & 0x1F)
+#define R(color) GET_CHAN(color, RED)
+#define G(color) GET_CHAN(color, GREEN)
+#define B(color) GET_CHAN(color, BLUE)
+
+/* 8 bit rounding constants */
+#define ROUND_UP 7
+#define ROUND_NEAREST 4
+#define ROUND_DOWN 0
+
+/* tuning parameters */
+#define SKIP_FRAME_THRESH 13
+#define START_ONE_COLOR_THRESH 8
+#define CONTINUE_ONE_COLOR_THRESH 0
+#define SIXTEEN_COLOR_THRESH 24
+
+//#define RPZA_DITHER
+
+/* debug modes */
+//#define DEBUG_SKIP
+//#define DEBUG_SIXTEEN
+//#define DEBUG_STATS
+
+typedef enum channel_offset {
+ RED = 2,
+ GREEN = 1,
+ BLUE = 0,
+} channel_offset;
+
+typedef struct BlockInfo {
+ int row;
+ int col;
+ int block_width;
+ int block_height;
+ int image_width;
+ int image_height;
+ int block_index;
+ uint16_t start;
+ int rowstride;
+ int blocks_per_row;
+ int total_blocks;
+} BlockInfo;
+
+
+static void get_colors(uint8_t *colorB, uint8_t *colorA, uint8_t color4[4][3])
+{
+ uint8_t step;
+
+ color4[0][0] = colorB[0];
+ color4[0][1] = colorB[1];
+ color4[0][2] = colorB[2];
+
+ color4[3][0] = colorA[0];
+ color4[3][1] = colorA[1];
+ color4[3][2] = colorA[2];
+
+ // red components
+ step = (color4[3][0] - color4[0][0] + 1) / 3;
+ color4[1][0] = color4[0][0] + step;
+ color4[2][0] = color4[3][0] - step;
+
+ // green components
+ step = (color4[3][1] - color4[0][1] + 1) / 3;
+ color4[1][1] = color4[0][1] + step;
+ color4[2][1] = color4[3][1] - step;
+
+ // blue components
+ step = (color4[3][2] - color4[0][2] + 1) / 3;
+ color4[1][2] = color4[0][2] + step;
+ color4[2][2] = color4[3][2] - step;
+}
+
+
+static int get_block_info(BlockInfo *bi, int block)
+/* Fill BlockInfo struct with information about a 4x4 block of the image */
+{
+ bi->row = block / bi->blocks_per_row;
+ bi->col = block % bi->blocks_per_row;
+
+ // test for right edge block
+ if (bi->col == bi->blocks_per_row - 1 && (bi->image_width % 4) != 0) {
+ bi->block_width = bi->image_width % 4;
+ } else {
+ bi->block_width = 4;
+ }
+
+ // test for bottom edge block
+ if (bi->row == (bi->image_height / 4) && (bi->image_height % 4) != 0) {
+ bi->block_height = bi->image_height % 4;
+ } else {
+ bi->block_height = 4;
+ }
+
+ return block ? (bi->col * 4 * PIXELSTRIDE) + (bi->row * bi->rowstride * 4) : 0;
+}
+
+
+static uint16_t round_rgb24_to_rgb555(uint8_t * rgb24, int bias)
+/*
+ * Round a 24 bit rgb value to a 15 bit rgb value. The bias parameter
+ * specifies the rounding direction.
+ */
+{
+ uint16_t rgb555 = 0;
+ uint32_t r, g, b;
+
+ r = (uint32_t)rgb24[0] + bias;
+ g = (uint32_t)rgb24[1] + bias;
+ b = (uint32_t)rgb24[2] + bias;
+
+ r = r / 8;
+ g = g / 8;
+ b = b / 8;
+
+ /* clamp 0-31 */
+ if (r > 31) {
+ r = 31;
+ }
+ if (g > 31) {
+ g = 31;
+ }
+ if (b > 31) {
+ b = 31;
+ }
+
+ rgb555 |= (r << 10);
+ rgb555 |= (g << 5);
+ rgb555 |= (b << 0);
+
+ return rgb555;
+}
+
+#ifdef RPZA_DITHER
+
+
+static uint16_t dither_rgb24_to_rgb555(uint8_t * rgb24)
+/*
+ * Dither a 24 bit rgb value to a 15 bit rgb value.
+ */
+{
+ uint16_t rgb555 = 0;
+ uint32_t r, g, b;
+
+ r = (uint32_t)rgb24[0] + (uint32_t)(random() % 8);
+ g = (uint32_t)rgb24[1] + (uint32_t)(random() % 8);
+ b = (uint32_t)rgb24[2] + (uint32_t)(random() % 8);
+
+ r = r / 8;
+ g = g / 8;
+ b = b / 8;
+
+ /* clamp 0-31 */
+ if (r > 31) {
+ r = 31;
+ }
+ if (g > 31) {
+ g = 31;
+ }
+ if (b > 31) {
+ b = 31;
+ }
+
+ rgb555 |= (r << 10);
+ rgb555 |= (g << 5);
+ rgb555 |= (b << 0);
+
+ return rgb555;
+}
+#endif
+
+
+static int diff_colors(uint8_t *colorA, uint8_t *colorB)
+/*
+ * Returns the total difference between two 24 bit color values
+ */
+{
+ int tot;
+ tot = SQR(colorA[0] - colorB[0]);
+ tot += SQR(colorA[1] - colorB[1]);
+ tot += SQR(colorA[2] - colorB[2]);
+ return tot;
+}
+
+static int max_component_diff(uint8_t *colorA, uint8_t *colorB)
+/*
+ * Returns the maximum channel difference between two 24 bit color values
+ */
+{
+ int i, diff, max = 0;
+
+ for (i = 0; i < 3; i++) {
+ diff = abs(colorA[i] - colorB[i]);
+ if (diff > max) {
+ max = diff;
+ }
+ }
+ return max;
+}
+
+#if 0
+static void print_rgb24_color(uint8_t *color)
+{
+ printf("r{%02d} g{%02d} b{%02d}", color[0], color[1], color[2]);
+}
+
+static void print_block(uint8_t *block, BlockInfo *bi)
+{
+ int x, y;
+
+ for (y = 0; y < bi->block_height; y++) {
+ for (x = 0; x < bi->block_width; x++) {
+ printf("%d: ", x);
+ print_rgb24_color(&block[x * PIXELSTRIDE]);
+ printf("\n");
+ }
+ printf("\n");
+ block += bi->rowstride;
+ }
+}
+#endif
+
+static void get_max_component_diff(BlockInfo *bi, uint8_t *block_ptr,
+ uint8_t *min, uint8_t *max, channel_offset *chan)
+/*
+ * Find the channel that has the largest difference between minimum and maximum
+ * color values. Put the minimum value in min, maximum in max and the channel
+ * in chan.
+ */
+{
+ int x, y;
+ uint8_t min_r, max_r, min_g, max_g, min_b, max_b;
+ uint8_t r, g, b;
+
+ // fix warning about uninitialized vars
+ min_r = min_g = min_b = UINT8_MAX;
+ max_r = max_g = max_b = 0;
+
+ // loop thru and compare pixels
+ for (y = 0; y < bi->block_height; y++) {
+ for (x = 0; x < bi->block_width; x++){
+ // TODO: optimize
+ min_r = MIN(block_ptr[(x * PIXELSTRIDE) + 2], min_r);
+ min_g = MIN(block_ptr[(x * PIXELSTRIDE) + 1], min_g);
+ min_b = MIN(block_ptr[(x * PIXELSTRIDE) + 0], min_b);
+
+ max_r = MAX(block_ptr[(x * PIXELSTRIDE) + 2], max_r);
+ max_g = MAX(block_ptr[(x * PIXELSTRIDE) + 1], max_g);
+ max_b = MAX(block_ptr[(x * PIXELSTRIDE) + 0], max_b);
+ }
+ block_ptr += bi->rowstride;
+ }
+
+ r = max_r - min_r;
+ g = max_g - min_g;
+ b = max_b - min_b;
+
+ if (r > g && r > b) {
+ *max = max_r;
+ *min = min_r;
+ *chan = RED;
+ } else if (g > b && g >= r) {
+ *max = max_g;
+ *min = min_g;
+ *chan = GREEN;
+ } else {
+ *max = max_b;
+ *min = min_b;
+ *chan = BLUE;
+ }
+}
+
+static int compare_blocks(uint8_t *block1, uint8_t *block2, BlockInfo *bi, int thresh)
+/*
+ * Compare two 4x4 blocks to determine if the total difference between the
+ * blocks is greater than the thresh parameter. Returns -1 if difference
+ * exceeds threshold or zero otherwise.
+ */
+{
+ int x, y, diff = 0;
+ for (y = 0; y < bi->block_height; y++) {
+ for (x = 0; x < bi->block_width; x++) {
+ diff = max_component_diff(&block1[x * PIXELSTRIDE], &block2[x * PIXELSTRIDE]);
+ if (diff >= thresh) {
+ return -1;
+ }
+ }
+ block1 += bi->rowstride;
+ block2 += bi->rowstride;
+ }
+ return 0;
+}
+
+
+static int leastsquares(uint8_t *block_ptr, BlockInfo *bi,
+ channel_offset xchannel, channel_offset ychannel,
+ double *slope, double *y_intercept, double *correlation_coef)
+/*
+ * Determine the fit of one channel to another within a 4x4 block. This
+ * is used to determine the best palette choices for 4-color encoding.
+ */
+{
+ double sumx = 0, sumy = 0, sumx2 = 0, sumy2 = 0, sumxy = 0,
+ sumx_sq = 0, sumy_sq = 0, tmp, tmp2;
+ int i, j, count;
+ uint8_t x, y;
+
+ count = bi->block_height * bi->block_width;
+
+ if (count < 2) {
+ return -1;
+ }
+
+ for (i = 0; i < bi->block_height; i++) {
+ for (j = 0; j < bi->block_width; j++){
+ x = block_ptr[j * PIXELSTRIDE + xchannel];
+ y = block_ptr[j * PIXELSTRIDE + ychannel];
+ sumx += x;
+ sumy += y;
+ sumx2 += x * x;
+ sumy2 += y * y;
+ sumxy += x * y;
+ }
+ block_ptr += bi->rowstride;
+ }
+
+
+ sumx_sq = sumx * sumx;
+ tmp = (count * sumx2 - sumx_sq);
+
+ // guard against div/0
+ if (tmp == 0) {
+ return -2;
+ }
+
+ sumy_sq = sumy * sumy;
+
+ *slope = (count * sumxy - sumx * sumy) / tmp;
+ *y_intercept = (sumy - (*slope) * sumx) / count;
+
+ tmp2 = count * sumy2 - sumy_sq;
+ if (tmp2 == 0) {
+ *correlation_coef = 0.0;
+ } else {
+ *correlation_coef = (count * sumxy - sumx * sumy) /
+ sqrt(tmp * tmp2);
+ }
+
+ return 0; // success
+}
+
+static int
+calc_lsq_max_fit_error(uint8_t *block_ptr, BlockInfo *bi,
+ int min, int max, int tmp_min, int tmp_max,
+ channel_offset xchannel, channel_offset ychannel)
+/*
+ * Determine the amount of error in the leastsquares fit.
+ */
+{
+ int i, j, x, y;
+ int err;
+ int max_err = 0;
+
+ for (i = 0; i < bi->block_height; i++) {
+ for (j = 0; j < bi->block_width; j++){
+ int x_inc, lin_y, lin_x;
+ x = block_ptr[j * PIXELSTRIDE + xchannel];
+ y = block_ptr[j * PIXELSTRIDE + ychannel];
+
+ /* calculate x_inc as the 4-color index (0..3) */
+ x_inc = floor( (x - min) * 3.0 / (max - min) + 0.5);
+ x_inc = MAX(MIN(3, x_inc), 0);
+
+ /* calculate lin_y corresponding to x_inc */
+ lin_y = (int)(tmp_min + (tmp_max - tmp_min) * x_inc / 3.0 + 0.5);
+
+ err = abs(lin_y - y);
+ if (err > max_err)
+ max_err = err;
+
+ /* calculate lin_x corresponding to x_inc */
+ lin_x = (int)(min + (max - min) * x_inc / 3.0 + 0.5);
+
+ err = abs(lin_x - x);
+ if (err > max_err)
+ max_err += err;
+ }
+ block_ptr += bi->rowstride;
+ }
+
+ return max_err;
+}
+
+static int match_color(uint8_t *color, uint8_t colors[4][3])
+/*
+ * Find the closest match to a color within the 4-color palette
+ */
+{
+ int ret = 0, channel, palette_entry, variance;
+ int smallest_variance = INT_MAX;
+ uint8_t dithered_color[3];
+ int range;
+
+
+ for (channel = 0; channel < 3; channel++) {
+ range = abs(colors[3][channel] - colors[0][channel]);
+
+#ifdef RPZA_DITHER
+ if (range > 0) {
+ int channel_value = color[channel] + ((random() % range) / 3.0) - (range / 6.0);
+ if (channel_value < 0) {
+ channel_value = 0;
+ }
+ if (channel_value > 255) {
+ channel_value = 255;
+ }
+ dithered_color[channel] = channel_value;
+ } else {
+#endif
+ dithered_color[channel] = color[channel];
+#ifdef RPZA_DITHER
+ }
+#endif
+ }
+
+ for (palette_entry = 0; palette_entry < 4; palette_entry++) {
+ variance = diff_colors(dithered_color, colors[palette_entry]);
+
+ if (variance < smallest_variance) {
+ smallest_variance = variance;
+ ret = palette_entry;
+ }
+ }
+ return ret;
+}
+
+
+static int encode_four_color_block(uint8_t *min_color, uint8_t *max_color,
+ PutBitContext *pb, uint8_t *block_ptr, BlockInfo *bi)
+/*
+ * Encode a block using the 4-color opcode and palette. return number of
+ * blocks encoded (until we implement multi-block 4 color runs this will
+ * always be 1)
+ */
+{
+ int x, y, idx;
+ uint8_t color4[4][3];
+ uint16_t rounded_max, rounded_min;
+
+ // round min and max wider
+ rounded_min = round_rgb24_to_rgb555(min_color, ROUND_DOWN);
+ rounded_max = round_rgb24_to_rgb555(max_color, ROUND_UP);
+
+ // put a and b colors
+ // encode 4 colors = first 16 bit color with MSB zeroed and...
+ put_bits(pb, 16, rounded_max & ~0x8000);
+ // ...second 16 bit color with MSB on.
+ put_bits(pb, 16, rounded_min | 0x8000);
+
+ // scale back up to 24 bit
+ min_color[0] = R(rounded_min) * 8;
+ min_color[1] = G(rounded_min) * 8;
+ min_color[2] = B(rounded_min) * 8;
+
+ max_color[0] = R(rounded_max) * 8;
+ max_color[1] = G(rounded_max) * 8;
+ max_color[2] = B(rounded_max) * 8;
+
+ get_colors(min_color, max_color, color4);
+
+ for (y = 0; y < 4; y++) {
+ for (x = 0; x < 4; x++) {
+ idx = match_color(&block_ptr[x * PIXELSTRIDE], color4);
+ put_bits(pb, 2, idx);
+ }
+ block_ptr += bi->rowstride;
+ }
+ return 1; // num blocks encoded
+}
+
+
+
+static void update_block_in_prev_frame(const uint8_t *src_pixels,
+ uint8_t *dest_pixels, const BlockInfo *bi, int block_counter)
+/*
+ * Copy a 4x4 block from the current frame buffer to the previous frame buffer.
+ */
+{
+ int y;
+
+ for (y = 0; y < 4; y++) {
+ memcpy (dest_pixels, src_pixels, 4 * PIXELSTRIDE);
+ dest_pixels += bi->rowstride;
+ src_pixels += bi->rowstride;
+ }
+}
+
+
+static int
+update_block_stats(BlockInfo *bi, uint8_t *block,
+ uint8_t min_color[3], uint8_t max_color[3],
+ int *total_rgb, int *total_pixels,
+ uint8_t avg_color[3], int first_block)
+/*
+ update_block_stats updates statistics for the specified block. If first_block,
+ it initializes the statistics. Otherwise it updates the statistics IF THIS
+ BLOCK IS SUITABLE TO CONTINUE A 1-COLOR RUN. That is, it checks whether
+ the range of colors (since the routine was called first_block != 0) are
+ all close enough intensities to be represented by a single color.
+
+ The routine returns 0 if this block is too different to be part of
+ the same run of 1-color blocks. The routine returns 1 if this
+ block can be part of the same 1-color block run.
+
+ If the routine returns 1, it also updates its arguments to include
+ the statistics of this block. Otherwise, the stats are unchanged
+ and don't include the current block.
+ */
+{
+ int x, y;
+ int is_in_range;
+ int total_pixels_blk;
+ int threshold;
+
+ uint8_t min_color_blk[3], max_color_blk[3];
+ int total_rgb_blk[3];
+ uint8_t avg_color_blk[3];
+
+ if (first_block) {
+ min_color[0] = UINT8_MAX;
+ min_color[1] = UINT8_MAX;
+ min_color[2] = UINT8_MAX;
+ max_color[0] = 0;
+ max_color[1] = 0;
+ max_color[2] = 0;
+ total_rgb[0] = 0;
+ total_rgb[1] = 0;
+ total_rgb[2] = 0;
+ *total_pixels = 0;
+ threshold = START_ONE_COLOR_THRESH;
+ } else {
+ threshold = CONTINUE_ONE_COLOR_THRESH;
+ }
+
+ /*
+ The *_blk variables will include the current block.
+ Initialize them based on the blocks so far.
+ */
+ min_color_blk[0] = min_color[0];
+ min_color_blk[1] = min_color[1];
+ min_color_blk[2] = min_color[2];
+ max_color_blk[0] = max_color[0];
+ max_color_blk[1] = max_color[1];
+ max_color_blk[2] = max_color[2];
+ total_rgb_blk[0] = total_rgb[0];
+ total_rgb_blk[1] = total_rgb[1];
+ total_rgb_blk[2] = total_rgb[2];
+ total_pixels_blk = *total_pixels + bi->block_height * bi->block_width;
+
+ /*
+ Update stats for this block's pixels
+ */
+ for (y = 0; y < bi->block_height; y++) {
+ for (x = 0; x < bi->block_width; x++) {
+
+ total_rgb_blk[0] += block[x * PIXELSTRIDE];
+ total_rgb_blk[1] += block[x * PIXELSTRIDE + 1];
+ total_rgb_blk[2] += block[x * PIXELSTRIDE + 2];
+
+ min_color_blk[0] = MIN(block[x * PIXELSTRIDE], min_color_blk[0]);
+ min_color_blk[1] = MIN(block[x * PIXELSTRIDE + 1], min_color_blk[1]);
+ min_color_blk[2] = MIN(block[x * PIXELSTRIDE + 2], min_color_blk[2]);
+
+ max_color_blk[0] = MAX(block[x * PIXELSTRIDE], max_color_blk[0]);
+ max_color_blk[1] = MAX(block[x * PIXELSTRIDE + 1], max_color_blk[1]);
+ max_color_blk[2] = MAX(block[x * PIXELSTRIDE + 2], max_color_blk[2]);
+ }
+ block += bi->rowstride;
+ }
+
+ /*
+ Calculate average color including current block.
+ */
+ avg_color_blk[0] = total_rgb_blk[0] / total_pixels_blk;
+ avg_color_blk[1] = total_rgb_blk[1] / total_pixels_blk;
+ avg_color_blk[2] = total_rgb_blk[2] / total_pixels_blk;
+
+ /*
+ Are all the pixels within threshold of the average color?
+ */
+ is_in_range = (max_color_blk[0] - avg_color_blk[0] <= threshold &&
+ max_color_blk[1] - avg_color_blk[1] <= threshold &&
+ max_color_blk[2] - avg_color_blk[2] <= threshold &&
+ avg_color_blk[0] - min_color_blk[0] <= threshold &&
+ avg_color_blk[1] - min_color_blk[1] <= threshold &&
+ avg_color_blk[2] - min_color_blk[2] <= threshold);
+
+ if (is_in_range) {
+ /*
+ Set the output variables to include this block.
+ */
+ min_color[0] = min_color_blk[0];
+ min_color[1] = min_color_blk[1];
+ min_color[2] = min_color_blk[2];
+ max_color[0] = max_color_blk[0];
+ max_color[1] = max_color_blk[1];
+ max_color[2] = max_color_blk[2];
+ total_rgb[0] = total_rgb_blk[0];
+ total_rgb[1] = total_rgb_blk[1];
+ total_rgb[2] = total_rgb_blk[2];
+ *total_pixels = total_pixels_blk;
+ avg_color[0] = avg_color_blk[0];
+ avg_color[1] = avg_color_blk[1];
+ avg_color[2] = avg_color_blk[2];
+ }
+
+ return is_in_range;
+}
+
+
+static void rpza_encode_stream(RpzaContext *s, AVFrame *pict)
+/*
+ * Encode a single frame and add it to the sequence
+ */
+{
+ BlockInfo bi;
+ int block_counter = 0;
+ int n_blocks;
+ int total_blocks;
+ int prev_block_offset;
+ int block_offset = 0;
+ uint8_t min = 0, max = 0;
+ channel_offset chan;
+ int i;
+ int tmp_min, tmp_max;
+ int total_rgb[3];
+ uint8_t avg_color[3];
+ int pixel_count;
+ uint8_t min_color[3], max_color[3];
+ double slope, y_intercept, correlation_coef;
+ uint8_t *src_pixels = (uint8_t*) pict->data[0];
+ uint8_t *prev_pixels = (uint8_t*) s->prev_frame.data[0];
+
+ /* Number of 4x4 blocks in frame. */
+ total_blocks = ((s->frame_width + 3) / 4) * ((s->frame_height + 3) / 4);
+
+ bi.image_width = s->frame_width;
+ bi.image_height = s->frame_height;
+ bi.rowstride = pict->linesize[0];
+
+ bi.blocks_per_row = (s->frame_width + 3) / 4;
+
+ while (block_counter < total_blocks) {
+
+ // SKIP CHECK
+ // make sure we have a valid previous frame and we're not writing
+ // a key frame
+ if (!s->first_frame && s->current_frame.pict_type == FF_P_TYPE) {
+
+ n_blocks = 0;
+ prev_block_offset = 0;
+
+ while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
+
+ block_offset = get_block_info(&bi, block_counter + n_blocks);
+
+ // multi-block opcodes cannot span multiple rows.
+ // If we're starting a new row, break out and write the opcode
+ /* TODO: Should eventually use bi.row here to determine when a
+ row break occurs, but that is currently breaking the
+ quicktime player. This is probably due to a bug in the
+ way I'm calculating the current row.
+ */
+ if (prev_block_offset && block_offset - prev_block_offset > 12) {
+ break;
+ }
+
+ prev_block_offset = block_offset;
+
+ if (compare_blocks(&prev_pixels[block_offset],
+ &src_pixels[block_offset], &bi, SKIP_FRAME_THRESH) != 0) {
+ // write out skipable blocks
+ if (n_blocks) {
+#ifdef DEBUG_STATS
+ s->skip_count++;
+ s->skip_blocks_count += n_blocks;
+#endif
+
+#ifdef DEBUG_SKIP
+ // write one color opcode.
+ put_bits(&s->pb, 8, 0xa0 | (n_blocks - 1));
+ // write color to encode.
+ put_bits(&s->pb, 16, 0x0FF0);
+#else
+ // write skip opcode
+ put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
+#endif
+ block_counter += n_blocks;
+
+ goto post_skip;
+ }
+ break;
+ }
+
+ /*
+ * NOTE: we don't update skipped blocks in the previous frame buffer
+ * since skipped needs always to be compared against the first skipped
+ * block to avoid artifacts during gradual fade in/outs.
+ */
+
+ // DEBUG PREV FRAME WRITING
+ // update_block_in_prev_frame(&src_pixels[block_offset],
+ // &prev_pixels[block_offset], &bi, block_counter + n_blocks);
+
+ n_blocks++;
+ }
+
+ // we're either at the end of the frame or we've reached the maximum
+ // of 32 blocks in a run. Write out the run.
+ if (n_blocks) {
+#ifdef DEBUG_STATS
+ s->skip_count++;
+ s->skip_blocks_count += n_blocks;
+#endif
+
+#ifdef DEBUG_SKIP
+ // write skip opcodes as one color opcodes for debugging
+ put_bits(&s->pb, 8, 0xa0 | (n_blocks - 1));
+ // write debug color to encode skips as.
+ put_bits(&s->pb, 16, 0x0FF0);
+#else
+ // write skip opcode
+ put_bits(&s->pb, 8, 0x80 | (n_blocks - 1));
+#endif
+ block_counter += n_blocks;
+
+ continue;
+ }
+
+ } else {
+ block_offset = get_block_info(&bi, block_counter);
+ }
+post_skip :
+
+ // ONE COLOR CHECK
+ if (update_block_stats(&bi, &src_pixels[block_offset],
+ min_color, max_color,
+ total_rgb, &pixel_count, avg_color, 1)) {
+ int first_block_offset;
+ first_block_offset = prev_block_offset = block_offset;
+
+ n_blocks = 1;
+
+ /* update this block in the previous frame buffer */
+ update_block_in_prev_frame(&src_pixels[block_offset],
+ &prev_pixels[block_offset], &bi, block_counter + n_blocks);
+
+ // check for subsequent blocks with the same color
+ while (n_blocks < 32 && block_counter + n_blocks < total_blocks) {
+ block_offset = get_block_info(&bi, block_counter + n_blocks);
+
+ // multi-block opcodes cannot span multiple rows.
+ // If we've hit end of a row, break out and write the opcode
+ if (block_offset - prev_block_offset > 12) {
+ break;
+ }
+
+ if (!update_block_stats(&bi, &src_pixels[block_offset],
+ min_color, max_color,
+ total_rgb, &pixel_count, avg_color, 0)) {
+ break;
+ }
+
+ prev_block_offset = block_offset;
+
+ /* update this block in the previous frame buffer */
+ update_block_in_prev_frame(&src_pixels[block_offset],
+ &prev_pixels[block_offset], &bi, block_counter + n_blocks);
+
+ n_blocks++;
+ }
+
+
+#ifdef DEBUG_STATS
+ s->one_count++;
+ s->one_blocks_count += n_blocks;
+#endif
+ // write one color opcode.
+ put_bits(&s->pb, 8, 0xa0 | (n_blocks - 1));
+ // write color to encode.
+ put_bits(&s->pb, 16, round_rgb24_to_rgb555(avg_color, ROUND_NEAREST));
+ // skip past the blocks we've just encoded.
+ block_counter += n_blocks;
+ } else { // FOUR COLOR CHECK
+ int err = 0;
+
+ // get max component diff for block
+ get_max_component_diff(&bi, &src_pixels[block_offset], &min, &max, &chan);
+
+ min_color[0] = 0;
+ max_color[0] = 0;
+ min_color[1] = 0;
+ max_color[1] = 0;
+ min_color[2] = 0;
+ max_color[2] = 0;
+
+ // run least squares against other two components
+ for (i = 0; i < 3; i++) {
+
+ if (i == chan) {
+ min_color[i] = min;
+ max_color[i] = max;
+ continue;
+ }
+
+ slope = y_intercept = correlation_coef = 0;
+
+ if (leastsquares(&src_pixels[block_offset], &bi, chan, i,
+ &slope, &y_intercept, &correlation_coef)) {
+ min_color[i] = src_pixels[block_offset + i];
+ max_color[i] = src_pixels[block_offset + i];
+
+ } else {
+ tmp_min = (int)(0.5 + min * slope + y_intercept);
+ tmp_max = (int)(0.5 + max * slope + y_intercept);
+
+ // clamp min and max color values
+ if (tmp_min > 255) {
+ tmp_min = 255;
+ }
+ if (tmp_max > 255) {
+ tmp_max = 255;
+ }
+ if (tmp_min < 0) {
+ tmp_min = 0;
+ }
+ if (tmp_max < 0) {
+ tmp_max = 0;
+ }
+
+ err = MAX(calc_lsq_max_fit_error(&src_pixels[block_offset], &bi,
+ min, max, tmp_min, tmp_max, chan, i), err);
+
+ min_color[i] = tmp_min;
+ max_color[i] = tmp_max;
+ }
+ }
+
+ if (err > SIXTEEN_COLOR_THRESH) { // DO SIXTEEN COLOR BLOCK
+ int x, y;
+ uint8_t *row_ptr;
+ uint16_t rgb555 = 0;
+
+#ifdef DEBUG_STATS
+ s->sixteen_count++;
+#endif
+
+#ifdef DEBUG_SIXTEEN
+ // write one color opcode.
+ put_bits(&s->pb, 8, 0xa0 | 0);
+ // write color to encode.
+ put_bits(&s->pb, 16, 0x0FF0);
+#else
+ block_offset = get_block_info(&bi, block_counter);
+
+ row_ptr = &src_pixels[block_offset];
+ // encode 16 colors = first 16 bit color with MSB zeroed and...
+#ifdef RPZA_DITHER
+ rgb555 = dither_rgb24_to_rgb555(row_ptr);
+#else
+ rgb555 = round_rgb24_to_rgb555(row_ptr, ROUND_NEAREST);
+#endif
+ put_bits(&s->pb, 16, rgb555 & ~0x8000);
+
+ row_ptr += PIXELSTRIDE;
+
+ // ...second 16 bit color with MSB zeroed.
+#ifdef RPZA_DITHER
+ rgb555 = dither_rgb24_to_rgb555(row_ptr);
+#else
+ rgb555 = round_rgb24_to_rgb555(row_ptr, ROUND_NEAREST);
+#endif
+ put_bits(&s->pb, 16, rgb555 & ~0x8000);
+
+ row_ptr += PIXELSTRIDE;
+
+ x = 2; // skip first two pixels we just encoded above.
+
+ for (y = 0; y < 4; y++) {
+ for (; x < 4; x++){
+#ifdef RPZA_DITHER
+ rgb555 = dither_rgb24_to_rgb555(row_ptr);
+#else
+ rgb555 = round_rgb24_to_rgb555(row_ptr, ROUND_NEAREST);
+#endif
+ put_bits(&s->pb, 16, rgb555);
+ row_ptr+= PIXELSTRIDE;
+ }
+ x = 0;
+ row_ptr += bi.rowstride - (4 * PIXELSTRIDE);
+ }
+#endif
+ block_counter++;
+
+ } else { // FOUR COLOR BLOCK
+#ifdef DEBUG_STATS
+ s->four_count++;
+#endif
+ block_counter += encode_four_color_block(min_color, max_color,
+ &s->pb, &src_pixels[block_offset], &bi);
+ }
+
+ /* update this block in the previous frame buffer */
+ update_block_in_prev_frame(&src_pixels[block_offset],
+ &prev_pixels[block_offset], &bi, block_counter);
+ }
+ }
+ return;
+}
+
+
+static int rpza_encode_init(AVCodecContext *avctx)
+/*
+ * Called by libavcodec before encoding begins
+ */
+{
+ RpzaContext * const s = avctx->priv_data;
+
+ dsputil_init(&s->dsp, avctx);
+
+ avctx->coded_frame = (AVFrame*)&s->current_frame;
+
+ avctx->coded_frame->pict_type = FF_I_TYPE;
+ avctx->coded_frame->key_frame = 1;
+
+ s->frame_width = avctx->width;
+ s->frame_height = avctx->height;
+
+ s->avctx = avctx;
+
+#ifdef DEBUG_STATS
+ /* init debug counters */
+ s->skip_count = 0;
+ s->skip_blocks_count = 0;
+ s->one_count = 0;
+ s->one_blocks_count = 0;
+ s->four_count = 0;
+ s->sixteen_count = 0;
+#endif
+ return 0;
+}
+
+
+#ifdef DEBUG_SKIP
+#include "../libavformat/avformat.h"
+static void dump_img_to_sgi(AVFrame *rgb_frame, int width, int height, char *filename)
+/*
+ * Dump frame to an sgi file for debuggimg
+ */
+{
+ int err;
+ AVImageFormat *image_fmt;
+ AVImageInfo img_info;
+ ByteIOContext pb;
+
+ for (image_fmt = first_image_format; image_fmt != NULL;
+ image_fmt = image_fmt->next) {
+ if (strncmp(image_fmt->name, "sgi", 3) == 0) {
+ break;
+ }
+ }
+ img_info.pict.data[0] = rgb_frame->data[0];
+ img_info.pict.linesize[0] = rgb_frame->linesize[0];
+ img_info.pix_fmt = PIX_FMT_RGB24;
+ img_info.width = width;
+ img_info.height = height;
+ img_info.interleaved = 0;
+
+ // open the file
+ err = url_fopen(&pb, filename, URL_RDWR);
+ if (err < 0) {
+ printf("Could not open %s %d\n", filename, err);
+ exit(1);
+ }
+
+ url_setbufsize(&pb, 4096);
+ image_fmt->img_write(&pb, &img_info);
+ url_fclose(&pb);
+}
+#endif
+
+static int rpza_encode_frame(AVCodecContext *avctx, unsigned char *buf,
+ int buf_size, void *data)
+/* Called by libavformat to encode a single frame into rpza. */
+{
+ RpzaContext * const s = avctx->priv_data;
+ int encoded_frame_size;
+ AVFrame *pict = data;
+ AVFrame * const p = (AVFrame*)&s->current_frame;
+
+ s->current_frame.reference = 1;
+ if(!s->current_frame.data[0]){
+ avctx->get_buffer(avctx, &s->current_frame);
+ }
+
+ init_put_bits(&s->pb, buf, buf_size);
+
+ // skip 4 byte header, write it later once the size of the chunk is known
+ put_bits(&s->pb, 32, 0x00);
+
+ *p = *pict;
+ p->pict_type = avctx->frame_number % avctx->gop_size ? FF_P_TYPE : FF_I_TYPE;
+ p->key_frame = p->pict_type == FF_I_TYPE;
+
+ if(!s->prev_frame.data[0]){
+ s->first_frame = 1;
+ avctx->get_buffer(avctx, &s->prev_frame);
+ } else {
+ s->first_frame = 0;
+ }
+
+ s->prev_frame.linesize[0] = pict->linesize[0];
+
+ rpza_encode_stream(s, p);
+ flush_put_bits(&s->pb);
+
+ encoded_frame_size = (put_bits_count(&s->pb) / 8);
+
+ // write header opcode
+ buf[0] = 0xe1; // chunk opcode
+
+ // write chunk length
+ buf[1] = encoded_frame_size >> 16;
+ buf[2] = encoded_frame_size >> 8;
+ buf[3] = encoded_frame_size;
+
+#ifdef DEBUG_SKIP
+ {
+ // Write out prev and current frame buffers to sgi
+ static int frame = 1;
+
+ char current[32], prev[32];
+ snprintf(current, 17, "current-%04d.sgi", frame);
+ snprintf(prev, 17, "prev-%04d.sgi", frame);
+ dump_img_to_sgi(&s->current_frame, s->frame_width, s->frame_height, current);
+ dump_img_to_sgi(&s->prev_frame, s->frame_width, s->frame_height, prev);
+
+ frame++;
+ }
+#endif
+
+ return encoded_frame_size;
+}
+
+
+static int rpza_encode_end(AVCodecContext *avctx)
+/* Called by libavformat when encoding is finished. */
+{
+ RpzaContext *s = (RpzaContext *)avctx->priv_data;
+
+#ifdef DEBUG_STATS
+ int total_blocks = s->skip_blocks_count + s->one_blocks_count + s->four_count + s->sixteen_count;
+ printf("\n\nRPZA DEBUG STATS\n");
+ printf("skip %d\n", s->skip_count);
+ printf("skip blocks %d (%0.2f%%)\n", s->skip_blocks_count, (float) s->skip_blocks_count / total_blocks * 100);
+ printf("one %d\n", s->one_count);
+ printf("one blocks %d (%0.2f%%)\n", s->one_blocks_count, (float) s->one_blocks_count / total_blocks * 100);
+ printf("four %d\n", s->four_count);
+ printf("four blocks %d (%0.2f%%)\n", s->four_count, (float) s->four_count / total_blocks * 100);
+ printf("sixteen %d\n", s->sixteen_count);
+ printf("sixteen blocks %d (%0.2f%%)\n", s->sixteen_count, (float) s->sixteen_count / total_blocks * 100);
+ printf("total blocks %d\n", total_blocks);
+#endif
+
+ if (s->current_frame.data[0]) {
+ avctx->release_buffer(avctx, &s->current_frame);
+ }
+
+ if (s->prev_frame.data[0]) {
+ avctx->release_buffer(avctx, &s->prev_frame);
+ }
+
+ return 0;
+}
+
+AVCodec rpza_encoder = {
+ "rpza",
+ CODEC_TYPE_VIDEO,
+ CODEC_ID_RPZA,
+ sizeof(RpzaContext),
+ rpza_encode_init,
+ rpza_encode_frame,
+ rpza_encode_end,
+ .pix_fmts= (enum PixelFormat[]){PIX_FMT_RGB24, -1},
+};
+
+#endif //CONFIG_ENCODERS
diff --git a/libavformat/movenc.c b/libavformat/movenc.c
index 20a9421..ae40e3b 100644
--- a/libavformat/movenc.c
+++ b/libavformat/movenc.c
@@ -497,6 +497,7 @@ const CodecTag codec_movvideo_tags[] = {
{ CODEC_ID_MPEG4, MKTAG('m', 'p', '4', 'v') },
{ CODEC_ID_H263, MKTAG('s', '2', '6', '3') },
{ CODEC_ID_DVVIDEO, MKTAG('d', 'v', 'c', ' ') },
+ { CODEC_ID_RPZA, MKTAG('r', 'p', 'z', 'a') },
{ 0, 0 },
};
--
1.7.1
--------------000806090108070003070803--
More information about the ffmpeg-devel
mailing list