Hi,
Aaron, thanks for your advice :)
I tried the different filtering. Using SWS_POINT produces a picture with
the value 253 for every byte, this is better, but still not optimal. To
me this seems like a bug? I rather expected that scaling down a complete
white image shall reproduce another complete white image...
I also wrote a small example to reproduce the issue:
//setup src and dst
int w = 1920;
int h = 1080;
int size = 3 * w * h;
unsigned char src[size];
memset(src,0xFF,size);
int new_w = 960;// 1920 960
int new_h = 540;// 1080 540
int new_size = 3 * new_w * new_h;
unsigned char dst[new_size];
//memset(dst, 0x00, new_size);
//struct SwsContext *sws_getCachedContext(struct SwsContext
*context, int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int
dstH, enum PixelFormat dstFormat, int flags, SwsFilter *srcFilter,
SwsFilter *dstFilter, const double *param);
m_convert_context = sws_getCachedContext(m_convert_context, w,
h, PIX_FMT_RGB24, new_w, new_h, PIX_FMT_RGB24,
SWS_FAST_BILINEAR/*SWS_POINT*/, NULL, NULL, NULL);
//use arrays of pointers, sws_scale need such an array, despite
its just one plane
unsigned char* src_data[3] = {src, NULL, NULL};
int src_stride[3] = {3*w, 0, 0}; //RGB stride is 3*width
//use offsets to set pointers
unsigned char* dst_data[3]= {dst, NULL, NULL};
int dst_stride[3] = {3*new_w, 0, 0};
//convert
//int sws_scale(struct SwsContext *context, const uint8_t*
const srcSlice[], const int srcStride[], int srcSliceY, int srcSliceH,
uint8_t* const dst[], const int dstStride[]);
int ret = sws_scale(m_convert_context, src_data, src_stride, 0,
h, dst_data, dst_stride);
if (ret < 0){std::cout << "sws_scale failed: " << ret <<
std::endl;}
else{std::cout << "sws_scale succeeded: " << ret << std::endl;}
//check the values
for(int x=0; x<new_size; x++){
if(dst[x]!= 0xFF){std::cout << "wrong value " << (int)
dst[x] << " at " << x << std::endl;}
}
Regards ... Johannes
On 06/29/2011 01:43 AM, Aaron Zinghini wrote:
Just a quick guess but I believe it is because of the filtering (SWS_FAST_BILINEAR).
Filtering is used to remove aliasing effects when resizing an image and therefore you
won't get a 1-1 mapping of pixels and the filtering will use neighbouring pixels to
basically "blend" or smooth the image so it's not blocky after resize.
If there's an option like SWS_NONE or maybe SWS_POINT then they might produce
more of a 1-1 mapping.
-----Original Message-----
From: [email protected] [mailto:[email protected]] On
Behalf Of Johannes Ebersold
Sent: Wednesday, 29 June 2011 1:54 AM
To: [email protected]
Subject: [Libav-user] Scaling via sws_scale() produces wrong colourvalues
Hi,
I'm currently trying to scale a RGB picture using libswscale.
I tested my application by creating a complete white image, set every
byte to 255. But in the output after the scaling every pixel got the
value "254 251 251". I don't know what i got wrong and would be thankful
for any advice.
My Code is part of a larger project, i will just paste a snippet here...
So far i think, that i got something wrong initializing or using the
swscale.
m_convert_context = sws_getCachedContext(m_convert_context,
src_res.first, src_res.second, PIX_FMT_RGB24, m_dst_res.first,
m_dst_res.second, PIX_FMT_RGB24, SWS_FAST_BILINEAR, NULL, NULL, NULL);
//src = Pointer to a packed RGB image with
unsigned char* src_data[3] = {src, NULL, NULL};
int src_stride[3] = {3*src_res.first, 0, 0}; //RGB stride is 3*width
/use offsets to set pointers, dst is just a pointer to some memory,
where the resulting image will be...
unsigned char* dst_data[3]= {dst, NULL, NULL};
int dst_stride[3] = {3*m_dst_res.first, 0, 0};
//convert
int ret = sws_scale(m_convert_context, src_data, src_stride, 0,
src_res.second, dst_data, dst_stride);
Thanks for any help!
Johannes
_______________________________________________
Libav-user mailing list
[email protected]
http://ffmpeg.org/mailman/listinfo/libav-user
_______________________________________________
Libav-user mailing list
[email protected]
http://ffmpeg.org/mailman/listinfo/libav-user
_______________________________________________
Libav-user mailing list
[email protected]
http://ffmpeg.org/mailman/listinfo/libav-user
