如何解决x264编码严重质量下降
我使用this回购将mjpeg流编码为h264,但输出效果不是很好。该流是iPhone的一系列屏幕截图。在输出流中,甚至设置应用程序中两个项目之间的线条也消失了。如何改善输出流质量? 这是x264-go用于初始化编码器的代码段
func NewEncoder(w io.Writer,opts *Options) (e *Encoder,err error) {
e = &Encoder{}
e.w = w
e.pts = 0
e.opts = opts
e.csp = x264c.CspI420
e.nals = make([]*x264c.Nal,3)
e.img = Newycbcr(image.Rect(0,e.opts.Width,e.opts.Height))
param := x264c.Param{}
if e.opts.Preset != "" && e.opts.Profile != "" {
ret := x264c.ParamDefaultPreset(¶m,e.opts.Preset,e.opts.Tune)
if ret < 0 {
err = fmt.Errorf("x264: invalid preset/tune name")
return
}
} else {
x264c.ParamDefault(¶m)
}
param.IWidth = int32(e.opts.Width)
param.IHeight = int32(e.opts.Height)
param.ICsp = e.csp
param.BVfrInput = 0
param.BRepeatHeaders = 1
param.BAnnexb = 1
param.ILogLevel = e.opts.LogLevel
if e.opts.FrameRate > 0 {
param.IFpsNum = uint32(e.opts.FrameRate)
param.IFpsDen = 1
param.IKeyintMax = int32(e.opts.FrameRate)
param.BIntraRefresh = 1
}
if e.opts.Profile != "" {
ret := x264c.ParamApplyProfile(¶m,e.opts.Profile)
if ret < 0 {
err = fmt.Errorf("x264: invalid profile name")
return
}
}
// Allocate on create instead while encoding
var picIn x264c.Picture
ret := x264c.PictureAlloc(&picIn,e.csp,int32(e.opts.Width),int32(e.opts.Height))
if ret < 0 {
err = fmt.Errorf("x264: cannot allocate picture")
return
}
e.picIn = picIn
defer func() {
// Cleanup if intialization fail
if err != nil {
x264c.PictureClean(&picIn)
}
}()
e.e = x264c.EncoderOpen(¶m)
if e.e == nil {
err = fmt.Errorf("x264: cannot open the encoder")
return
}
ret = x264c.EncoderHeaders(e.e,e.nals,&e.nnals)
if ret < 0 {
err = fmt.Errorf("x264: cannot encode headers")
return
}
if ret > 0 {
b := C.GoBytes(e.nals[0].PPayload,C.int(ret))
n,er := e.w.Write(b)
if er != nil {
err = er
return
}
if int(ret) != n {
err = fmt.Errorf("x264: error writing headers,size=%d,n=%d",ret,n)
}
}
return
}
,编码器选项定义为
opts := &x264.Options{
Width: int(width)/2*2,Height: int(height)/2*2,FrameRate: 15,Tune: "zerolatency",Preset: "medium",Profile: "baseline",LogLevel: x264.LogNone,}
我还使用了ffmpeg(尽管api很老了)
package screencast
import (
/*
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
#include <libavutil/avutil.h>
#include <libavutil/opt.h>
#include <libavutil/channel_layout.h>
#include <libavutil/common.h>
#include <libavutil/imgutils.h>
#include <libavutil/mathematics.h>
#include <libavutil/samplefmt.h>
typedef struct {
int w,h;
int pixfmt;
char *preset[2];
char *profile;
int bitrate;
int got;
AVCodec *c;
AVCodecContext *ctx;
AVFrame *f;
AVPacket pkt;
} h264enc_t;
static int h264enc_new(h264enc_t *m) {
m->c = avcodec_find_encoder(AV_CODEC_ID_H264);
m->ctx = avcodec_alloc_context3(m->c);
m->ctx->width = m->w;
m->ctx->height = m->h;
m->ctx->pix_fmt = m->pixfmt;
m->ctx->time_base = (AVRational){1,10};
av_opt_set(m->ctx->priv_data,"preset","slow",0);
av_opt_set(m->ctx->priv_data,"tune","zerolatency","profile","baseline","crf","18.0.",0);
m->f = av_frame_alloc();
m->f->format = m->ctx->pix_fmt;
m->f->width = m->ctx->width;
m->f->height = m->ctx->height;
avcodec_open2(m->ctx,m->c,NULL);
return av_image_alloc(m->f->data,m->f->linesize,m->ctx->width,m->ctx->height,m->ctx->pix_fmt,32);
}
*/
"C"
"errors"
"image"
"unsafe"
//"log"
)
type H264Encoder struct {
m C.h264enc_t
Header []byte
Pixfmt image.ycbcrSubsampleRatio
W,H int
pts int
}
func NewH264Encoder(w,h int) (m *H264Encoder,err error) {
m = &H264Encoder{}
m.m.w = (C.int)(w)
m.m.h = (C.int)(h)
m.W = w
m.H = h
m.Pixfmt = image.ycbcrSubsampleRatio420
m.m.pixfmt = C.AV_PIX_FMT_YUV420P
r := C.h264enc_new(&m.m)
if int(r) < 0 {
err = errors.New("open encoder Failed")
return
}
return
}
func (m *H264Encoder) Encode(img *image.ycbcr) (data []byte,err error) {
var f *C.AVFrame
if img == nil {
f = nil
} else {
if img.SubsampleRatio != m.Pixfmt {
err = errors.New("image pixfmt not match")
return
}
if img.Rect.Dx() != m.W || img.Rect.Dy() != m.H {
err = errors.New("image size not match")
return
}
f = m.m.f
f.data[0] = (*C.uint8_t)(unsafe.Pointer(&img.Y[0]))
f.data[1] = (*C.uint8_t)(unsafe.Pointer(&img.Cb[0]))
f.data[2] = (*C.uint8_t)(unsafe.Pointer(&img.Cr[0]))
f.linesize[0] = (C.int)(img.YStride)
f.linesize[1] = (C.int)(img.CStride)
f.linesize[2] = (C.int)(img.CStride)
}
C.av_init_packet(&m.m.pkt)
m.m.pkt.data = nil
m.m.pkt.size = 0
f.pts = (C.longlong)(m.pts)
m.pts++
r := C.avcodec_encode_video2(m.m.ctx,&m.m.pkt,f,&m.m.got)
defer C.av_packet_unref(&m.m.pkt)
if int(r) < 0 {
err = errors.New("encode Failed")
return
}
if m.m.got == 0 {
err = errors.New("no picture")
return
}
if m.m.pkt.size == 0 {
err = errors.New("packet size == 0")
return
}
data = make([]byte,m.m.pkt.size)
C.memcpy(
unsafe.Pointer(&data[0]),unsafe.Pointer(m.m.pkt.data),(C.size_t)(m.m.pkt.size),)
return data,nil
}
,但输出相同。但是,当我使用ffmpeg二进制文件时,结果非常好,所以我猜我设置了错误的参数,但是我不知道哪个。 任何建议,将不胜感激。如果您有更好的方法来实现这一目标,我将不胜感激。
供您参考:我必须使用golang进行此操作。
解决方法
事实证明这不是细节损失,真正的问题是颜色损失。输出流应为yuvj420p [0-255]格式,而不是yuv420p [16-239]。
版权声明:本文内容由互联网用户自发贡献,该文观点与技术仅代表作者本人。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容, 请发送邮件至 dio@foxmail.com 举报,一经查实,本站将立刻删除。