Skip to content

Commit

Permalink
move image transcode to lib file
Browse files Browse the repository at this point in the history 
since it was janky and awful to debug to have almost 300 lines of code in both the image and video transcoder, that function is now moved to its own lib file where it accepts any data type (just make sure it has a valid [] operator)
  • Loading branch information
@Like4Schnitzel
Like4Schnitzel committed May 31, 2024
1 parent d6e04ca commit f978f1d
Show file tree
Hide file tree
Showing 9 changed files with 318 additions and 527 deletions.
6 changes: 3 additions & 3 deletions src/CMakeLists.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ project( video-to-terminal )
find_package( OpenCV REQUIRED )
include_directories( ${OpenCV_INCLUDE_DIRS} )

add_executable( transcoder transcoder/main.cpp transcoder/videotranscoder.cpp libs/binaryutils.cpp libs/variousutils.cpp )
add_executable( transcoder transcoder/main.cpp transcoder/videotranscoder.cpp libs/binaryutils.cpp libs/variousutils.cpp libs/img2cimat.cpp )
#target_compile_options( transcoder PRIVATE -pg )
#target_link_options( transcoder PRIVATE -pg )
target_link_libraries( transcoder stdc++fs ${OpenCV_LIBS} )
Expand All @@ -15,7 +15,7 @@ add_executable( decoder decoder/main.cpp decoder/vtdidecoder.cpp libs/binaryutil
#target_link_options( decoder PRIVATE -pg -fsanitize=address )
target_link_libraries( decoder stdc++fs )

add_executable( img-viewer img-viewer/main.cpp img-viewer/imgviewer.cpp libs/variousutils.cpp )
add_executable( img-viewer img-viewer/main.cpp img-viewer/imgviewer.cpp libs/variousutils.cpp libs/img2cimat.cpp )
#target_compile_options( img-viewer PRIVATE -pg -fsanitize=address )
#target_link_options( img-viewer PRIVATE -pg -fsanitize=address )
target_link_libraries( img-viewer stdc++fs ${OpenCV_LIBS} )
Expand All @@ -24,6 +24,6 @@ add_executable (
media-viewer media-viewer/main.cpp media-viewer/mediaviewer.cpp
decoder/vtdidecoder.cpp
img-viewer/imgviewer.cpp
libs/termutils.cpp libs/variousutils.cpp libs/binaryutils.cpp libs/_kbhit.cpp
libs/termutils.cpp libs/variousutils.cpp libs/binaryutils.cpp libs/_kbhit.cpp libs/img2cimat.cpp
)
target_link_libraries( media-viewer stdc++fs ${OpenCV_LIBS} )
264 changes: 2 additions & 262 deletions src/img-viewer/imgviewer.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,273 +7,13 @@ ImgViewer::ImgViewer(const std::string path)
file = cv::imread(path);
}

int getColorDiff(cv::Scalar col1, cv::Scalar col2)
{
return pow(col1[0] - col2[0], 2) +
pow(col1[1] - col2[1], 2) +
pow(col1[2] - col2[2], 2);
}

CharInfo findBestBlockCharacter(cv::Mat img)
{
CharInfo maxDiffCharInfo;
int maxDiff = -1;
const int imageHeight = img.size().height;
const int imageWidth = img.size().width;
int currentOption;
cv::Mat fgRect, bgRect;

// skip upper half (0) since lower half can be used
// loop through the height characters from 1/8th to 7/8ths block
const double eigthHeight = imageHeight / 8.0;
double currentHeight = imageHeight;
// don't need to check full block (8) since it'll just go with fg=bg at check 1
for (currentOption = 1; currentOption < 8; currentOption++)
{
currentHeight -= eigthHeight;

fgRect = img(cv::Rect(0, round(currentHeight), imageWidth, imageHeight-round(currentHeight)));
if (fgRect.rows == 0) continue;
cv::Scalar avrgFgBGR = cv::mean(fgRect);

bgRect = img(cv::Rect(0, 0, imageWidth, round(currentHeight)));
if (bgRect.rows == 0) continue;
cv::Scalar avrgBgBGR = cv::mean(bgRect);

int colorDiff = getColorDiff(avrgFgBGR, avrgBgBGR);
if (colorDiff > maxDiff)
{
maxDiff = colorDiff;
for (int i = 0; i < 3; i++)
{
// turn around BGR so RGB gets saved instead
maxDiffCharInfo.foregroundRGB[i] = avrgFgBGR[2-i];
maxDiffCharInfo.backgroundRGB[i] = avrgBgBGR[2-i];
}
maxDiffCharInfo.chara = currentOption;

// no need to keep checking if the whole img is one color
if (colorDiff == 0)
{
return maxDiffCharInfo;
}
}
}

// loop through the width characters from 7/8ths to 1/8th
const double eigthWidth = imageWidth / 8.0;
double currentWidth = imageWidth;
for (currentOption = 9; currentOption < 0xF; currentOption++)
{
currentWidth -= eigthWidth;

fgRect = img(cv::Rect(0, 0, round(currentWidth), imageHeight));
if (fgRect.rows == 0) continue;
cv::Scalar avrgFgBGR = cv::mean(fgRect);

bgRect = img(cv::Rect(round(currentWidth), 0, imageWidth - round(currentWidth), imageHeight));
if (bgRect.rows == 0) continue;
cv::Scalar avrgBgBGR = cv::mean(bgRect);

int colorDiff = getColorDiff(avrgFgBGR, avrgBgBGR);
if (colorDiff > maxDiff)
{
maxDiff = colorDiff;
for (int i = 0; i < 3; i++)
{
// turn around BGR so RGB gets saved instead
maxDiffCharInfo.foregroundRGB[i] = avrgFgBGR[2-i];
maxDiffCharInfo.backgroundRGB[i] = avrgBgBGR[2-i];
}
maxDiffCharInfo.chara = currentOption;
}
}

// corner characters. They're weird.
// we're saving each of the means so that we can use them for calculations later
const int halfWidth = imageWidth / 2;
const int halfHeight = imageHeight / 2;
if (halfHeight > 0 && halfWidth > 0)
{
const int cornerArea = halfHeight * halfWidth;

fgRect = img(cv::Rect(0, halfHeight, halfWidth, halfHeight));
cv::Scalar lowLeftMean = cv::mean(fgRect);

fgRect = img(cv::Rect(halfWidth, halfHeight, halfWidth, halfHeight));
cv::Scalar lowRightMean = cv::mean(fgRect);

fgRect = img(cv::Rect(0, 0, halfWidth, halfHeight));
cv::Scalar upLeftMean = cv::mean(fgRect);

fgRect = img(cv::Rect(halfWidth, 0, halfWidth, halfHeight));
cv::Scalar upRightMean = cv::mean(fgRect);

cv::Scalar fgMean;
cv::Scalar bgMean;

const auto checkDiff = [&maxDiff, &maxDiffCharInfo, &fgMean, &bgMean, &currentOption]()
{
int colorDiff = getColorDiff(fgMean, bgMean);
if (colorDiff > maxDiff)
{
maxDiff = colorDiff;
for (int i = 0; i < 3; i++)
{
// turn around BGR so RGB gets saved instead
maxDiffCharInfo.foregroundRGB[i] = fgMean[2-i];
maxDiffCharInfo.backgroundRGB[i] = bgMean[2-i];
}
maxDiffCharInfo.chara = currentOption;
}
};

currentOption = 0x16;
fgMean = lowLeftMean;
bgMean = (lowRightMean + upLeftMean + upRightMean) / 3;
checkDiff();

currentOption = 0x17;
fgMean = lowRightMean;
bgMean = (lowLeftMean + upLeftMean + upRightMean) / 3;
checkDiff();

currentOption = 0x18;
fgMean = upLeftMean;
bgMean = (lowLeftMean + lowRightMean + upRightMean) / 3;
checkDiff();

// I don't fucking know why it just skips 4 elements until it does the last corner block but that's how it is.
currentOption = 0x1D;
fgMean = upRightMean;
bgMean = (lowLeftMean + lowRightMean + upLeftMean) / 3;
checkDiff();

// And now the double corner character.
currentOption = 0x1A;
fgMean = (upLeftMean + lowRightMean) / 2;
bgMean = (lowLeftMean + upRightMean) / 2;
checkDiff();
}

return maxDiffCharInfo;
}

double min(double a, double b) {if (a < b) return a; else return b;}
double max(double a, double b) {if (a < b) return b; else return a;}

/// @brief Calculates the mean color of a possibly pixel-misaligned submatrix inside a pixel matrix.
cv::Vec3b meanColWithSubPixels(cv::Mat mat, double x, double y, double width, double height)
{
// thanks to https://github.com/marceldobehere/ for helping me out with this one
cv::Vec3d result = {0, 0, 0};
double pixelCount = 0;

int x0 = (int)floor(x);
int y0 = (int)floor(y);
int x1 = (int)ceil(x+width);
int y1 = (int)ceil(y+height);

double ax0 = x;
double ay0 = y;
double ax1 = x + width;
double ay1 = y + height;

double x0Percent = ax0 - x0;
double y0Percent = ay0 - y0;
double x1Percent = x1 - ax1;
double y1Percent = y1 - ay1;

int maxW = min(x1 - x0 + 1, mat.cols - x0 - 1);
int maxH = min(y1 - y0 + 1, mat.rows - y0 - 1);
auto slice = mat(cv::Rect(x0, y0, maxW, maxH));

for (int y = 0; y < maxH; y++)
for (int x = 0; x < maxW; x++)
{
double pixelPercent = 1.0;
if (x == 0)
pixelPercent *= x0Percent;
if (x == maxW - 1)
pixelPercent *= x1Percent;
if (y == 0)
pixelPercent *= y0Percent;
if (y == maxH - 1)
pixelPercent *= y1Percent;

auto pixel = slice.at<cv::Vec3b>(x, y);
result += pixel * pixelPercent;
pixelCount += pixelPercent;
}



result /= pixelCount;
return (cv::Vec3b)result;
}

void ImgViewer::transcode(const int width, const int height)
{
this->width = width;
this->height = height;
transcodedFile.resize(width*height);

const int subPixelMatrixSize = 8;

const double widthPixelsPerChar = (double) file.size().width / this->width;
const double nthWidth = widthPixelsPerChar / subPixelMatrixSize;
const double heightPixelsPerChar = (double) file.size().height / this->height;
const double nthHeight = heightPixelsPerChar / subPixelMatrixSize;

int charIndex = 0;
std::vector<std::thread> threads;
threads.reserve(this->height);

double y = 0;
for (int i = 0; i < this->height; i++)
{
threads.emplace_back([&, y, charIndex](){
int ciIndexCopy = charIndex;
double xAnchor = 0;

for (int j = 0; j < this->width; j++)
{
double tempY = y;
// now we're gonna make an 8x8 pixel matrix to pass since we at max have an eighth of a character as resolution
cv::Mat3b rect(subPixelMatrixSize, subPixelMatrixSize);
for (int u = 0; u < subPixelMatrixSize; u++)
{
double tempX = xAnchor;
for (int v = 0; v < subPixelMatrixSize; v++)
{
rect.at<cv::Vec3b>(u, v) = meanColWithSubPixels(file, tempX, tempY, nthWidth, nthHeight);
tempX += nthWidth;
}
tempY += nthHeight;
}

CharInfo best = findBestBlockCharacter(rect);

for (int k = 0; k < 3; k++)
{
transcodedFile[ciIndexCopy].foregroundRGB[k] = best.foregroundRGB[k];
transcodedFile[ciIndexCopy].backgroundRGB[k] = best.backgroundRGB[k];
}
transcodedFile[ciIndexCopy].chara = best.chara;

xAnchor += widthPixelsPerChar;
ciIndexCopy++;
}
});
y += heightPixelsPerChar;
charIndex += this->width;
}

// wait for all threads to finish
for (auto& thread : threads)
{
thread.join();
}
transcodedFile.resize(width*height);
imgToCIMat(this->file, width, height, this->transcodedFile);
}

void ImgViewer::print()
Expand Down
1 change: 1 addition & 0 deletions src/img-viewer/imgviewer.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
#include <math.h>
#include "../libs/variousutils.hpp"
#include "../libs/charinfo.hpp"
#include "../libs/img2cimat.hpp"

namespace vtt {

Expand Down
Loading

0 comments on commit f978f1d

Please sign in to comment.