Walkthrough: Creating an Image-Processing Network

Article
08/03/2021

This document demonstrates how to create a network of asynchronous message blocks that perform image processing.

The network determines which operations to perform on an image on the basis of its characteristics. This example uses the dataflow model to route images through the network. In the dataflow model, independent components of a program communicate with one another by sending messages. When a component receives a message, it can perform some action and then pass the result of that action to another component. Compare this with the control flow model, in which an application uses control structures, for example, conditional statements, loops, and so on, to control the order of operations in a program.

A network that is based on dataflow creates a pipeline of tasks. Each stage of the pipeline concurrently performs part of the overall task. An analogy to this is an assembly line for automobile manufacturing. As each vehicle passes through the assembly line, one station assembles the frame, another installs the engine, and so on. By enabling multiple vehicles to be assembled simultaneously, the assembly line provides better throughput than assembling complete vehicles one at a time.

Prerequisites

Read the following documents before you start this walkthrough:

We also recommend that you understand the basics of GDI+ before you start this walkthrough.

Sections

This walkthrough contains the following sections:

Defining Image Processing Functionality
Creating the Image Processing Network
The Complete Example

Defining Image Processing Functionality

This section shows the support functions that the image processing network uses to work with images that are read from disk.

The following functions, GetRGB and MakeColor, extract and combine the individual components of the given color, respectively.

// Retrieves the red, green, and blue components from the given
// color value.
void GetRGB(DWORD color, BYTE& r, BYTE& g, BYTE& b)
{
   r = static_cast<BYTE>((color & 0x00ff0000) >> 16);
   g = static_cast<BYTE>((color & 0x0000ff00) >> 8);
   b = static_cast<BYTE>((color & 0x000000ff));
}

// Creates a single color value from the provided red, green, 
// and blue components.
DWORD MakeColor(BYTE r, BYTE g, BYTE b)
{
   return (r<<16) | (g<<8) | (b);
}

The following function, ProcessImage, calls the given std::function object to transform the color value of each pixel in a GDI+ Bitmap object. The ProcessImage function uses the concurrency::parallel_for algorithm to process each row of the bitmap in parallel.

// Calls the provided function for each pixel in a Bitmap object.
void ProcessImage(Bitmap* bmp, const function<void (DWORD&)>& f)
{
   int width = bmp->GetWidth();
   int height = bmp->GetHeight();

   // Lock the bitmap.
   BitmapData bitmapData;
   Rect rect(0, 0, bmp->GetWidth(), bmp->GetHeight());
   bmp->LockBits(&rect, ImageLockModeWrite, PixelFormat32bppRGB, &bitmapData);

   // Get a pointer to the bitmap data.
   DWORD* image_bits = (DWORD*)bitmapData.Scan0;

   // Call the function for each pixel in the image.
   parallel_for (0, height, [&, width](int y)
   {      
      for (int x = 0; x < width; ++x)
      {
         // Get the current pixel value.
         DWORD* curr_pixel = image_bits + (y * width) + x;

         // Call the function.
         f(*curr_pixel);
      }
   });

   // Unlock the bitmap.
   bmp->UnlockBits(&bitmapData);
}

The following functions, Grayscale, Sepiatone, ColorMask, and Darken, call the ProcessImage function to transform the color value of each pixel in a Bitmap object. Each of these functions uses a lambda expression to define the color transformation of one pixel.

// Converts the given image to grayscale.
Bitmap* Grayscale(Bitmap* bmp) 
{
   ProcessImage(bmp, 
      [](DWORD& color) {
         BYTE r, g, b;
         GetRGB(color, r, g, b);

         // Set each color component to the average of 
         // the original components.
         BYTE c = (static_cast<WORD>(r) + g + b) / 3;
         color = MakeColor(c, c, c);
      }
   );
   return bmp;
}

// Applies sepia toning to the provided image.
Bitmap* Sepiatone(Bitmap* bmp) 
{
   ProcessImage(bmp, 
      [](DWORD& color) {
         BYTE r0, g0, b0;
         GetRGB(color, r0, g0, b0);

         WORD r1 = static_cast<WORD>((r0 * .393) + (g0 *.769) + (b0 * .189));
         WORD g1 = static_cast<WORD>((r0 * .349) + (g0 *.686) + (b0 * .168));
         WORD b1 = static_cast<WORD>((r0 * .272) + (g0 *.534) + (b0 * .131));

         color = MakeColor(min(0xff, r1), min(0xff, g1), min(0xff, b1));
      }
   );
   return bmp;
}

// Applies the given color mask to each pixel in the provided image.
Bitmap* ColorMask(Bitmap* bmp, DWORD mask)
{
   ProcessImage(bmp, 
      [mask](DWORD& color) {
         color = color & mask;
      }
   );
   return bmp;
}

// Darkens the provided image by the given amount.
Bitmap* Darken(Bitmap* bmp, unsigned int percent)
{
   if (percent > 100)
      throw invalid_argument("Darken: percent must less than 100.");

   double factor = percent / 100.0;

   ProcessImage(bmp, 
      [factor](DWORD& color) {
         BYTE r, g, b;
         GetRGB(color, r, g, b);
         r = static_cast<BYTE>(factor*r);
         g = static_cast<BYTE>(factor*g);
         b = static_cast<BYTE>(factor*b);
         color = MakeColor(r, g, b);
      }
   );
   return bmp;
}

The following function, GetColorDominance, also calls the ProcessImage function. However, instead of changing the value of each color, this function uses concurrency::combinable objects to compute whether the red, green, or blue color component dominates the image.

// Determines which color component (red, green, or blue) is most dominant
// in the given image and returns a corresponding color mask.
DWORD GetColorDominance(Bitmap* bmp)
{
   // The ProcessImage function processes the image in parallel.
   // The following combinable objects enable the callback function
   // to increment the color counts without using a lock.
   combinable<unsigned int> reds;
   combinable<unsigned int> greens;
   combinable<unsigned int> blues;

   ProcessImage(bmp, 
      [&](DWORD& color) {
         BYTE r, g, b;
         GetRGB(color, r, g, b);
         if (r >= g && r >= b)
            reds.local()++;
         else if (g >= r && g >= b)
            greens.local()++;
         else
            blues.local()++;
      }
   );
   
   // Determine which color is dominant and return the corresponding
   // color mask.

   unsigned int r = reds.combine(plus<unsigned int>());
   unsigned int g = greens.combine(plus<unsigned int>());
   unsigned int b = blues.combine(plus<unsigned int>());

   if (r + r >= g + b)
      return 0x00ff0000;
   else if (g + g >= r + b)
      return 0x0000ff00;
   else
      return 0x000000ff;
}

The following function, GetEncoderClsid, retrieves the class identifier for the given MIME type of an encoder. The application uses this function to retrieve the encoder for a bitmap.

// Retrieves the class identifier for the given MIME type of an encoder.
int GetEncoderClsid(const WCHAR* format, CLSID* pClsid)
{
   UINT  num = 0;          // number of image encoders
   UINT  size = 0;         // size of the image encoder array in bytes

   ImageCodecInfo* pImageCodecInfo = nullptr;

   GetImageEncodersSize(&num, &size);
   if(size == 0)
      return -1;  // Failure

   pImageCodecInfo = (ImageCodecInfo*)(malloc(size));
   if(pImageCodecInfo == nullptr)
      return -1;  // Failure

   GetImageEncoders(num, size, pImageCodecInfo);

   for(UINT j = 0; j < num; ++j)
   {
      if( wcscmp(pImageCodecInfo[j].MimeType, format) == 0 )
      {
         *pClsid = pImageCodecInfo[j].Clsid;
         free(pImageCodecInfo);
         return j;  // Success
      }    
   }

   free(pImageCodecInfo);
   return -1;  // Failure
}

[Top]

Creating the Image Processing Network

This section describes how to create a network of asynchronous message blocks that perform image processing on every JPEG (.jpg) image in a given directory. The network performs the following image-processing operations:

For any image that is authored by Tom, convert to grayscale.
For any image that has red as the dominant color, remove the green and blue components and then darken it.
For any other image, apply sepia toning.

The network applies only the first image-processing operation that matches one of these conditions. For example, if an image is authored by Tom and has red as its dominant color, the image is only converted to grayscale.

After the network performs each image-processing operation, it saves the image to disk as a bitmap (.bmp) file.

The following steps show how to create a function that implements this image processing network and applies that network to every JPEG image in a given directory.

To create the image processing network

Create a function, ProcessImages, that takes the name of a directory on disk.
```
void ProcessImages(const wstring& directory)
{
}
```

In the ProcessImages function, create a countdown_event variable. The countdown_event class is shown later in this walkthrough.

// Holds the number of active image processing operations and 
// signals to the main thread that processing is complete.
countdown_event active(0);

Create a std::map object that associates a Bitmap object with its original file name.

// Maps Bitmap objects to their original file names.
map<Bitmap*, wstring> bitmap_file_names;

Add the following code to define the members of the image-processing network.

 //
 // Create the nodes of the network.
 //

 // Loads Bitmap objects from disk.
 transformer<wstring, Bitmap*> load_bitmap(
    [&](wstring file_name) -> Bitmap* {
       Bitmap* bmp = new Bitmap(file_name.c_str());
       if (bmp != nullptr)
          bitmap_file_names.insert(make_pair(bmp, file_name));
       return bmp;
    }
 );

 // Holds loaded Bitmap objects.
 unbounded_buffer<Bitmap*> loaded_bitmaps;

 // Converts images that are authored by Tom to grayscale.
 transformer<Bitmap*, Bitmap*> grayscale(
    [](Bitmap* bmp) {
       return Grayscale(bmp);
    },
    nullptr,
    [](Bitmap* bmp) -> bool {
       if (bmp == nullptr)
          return false;

       // Retrieve the artist name from metadata.
       UINT size = bmp->GetPropertyItemSize(PropertyTagArtist);
       if (size == 0)
          // Image does not have the Artist property.
          return false;

       PropertyItem* artistProperty = (PropertyItem*) malloc(size);
       bmp->GetPropertyItem(PropertyTagArtist, size, artistProperty);
       string artist(reinterpret_cast<char*>(artistProperty->value));
       free(artistProperty);
       
       return (artist.find("Tom ") == 0);
    }
 );
 
 // Removes the green and blue color components from images that have red as
 // their dominant color.
 transformer<Bitmap*, Bitmap*> colormask(
    [](Bitmap* bmp) {
       return ColorMask(bmp, 0x00ff0000);
    },
    nullptr,
    [](Bitmap* bmp) -> bool { 
       if (bmp == nullptr)
          return false;
       return (GetColorDominance(bmp) == 0x00ff0000);
    }
 );

 // Darkens the color of the provided Bitmap object.
 transformer<Bitmap*, Bitmap*> darken([](Bitmap* bmp) {
    return Darken(bmp, 50);
 });

 // Applies sepia toning to the remaining images.
 transformer<Bitmap*, Bitmap*> sepiatone(
    [](Bitmap* bmp) {
       return Sepiatone(bmp);
    },
    nullptr,
    [](Bitmap* bmp) -> bool { return bmp != nullptr; }
 );

 // Saves Bitmap objects to disk.
 transformer<Bitmap*, Bitmap*> save_bitmap([&](Bitmap* bmp) -> Bitmap* {
    // Replace the file extension with .bmp.
    wstring file_name = bitmap_file_names[bmp];
    file_name.replace(file_name.rfind(L'.') + 1, 3, L"bmp");
    
    // Save the processed image.
    CLSID bmpClsid;
    GetEncoderClsid(L"image/bmp", &bmpClsid);      
    bmp->Save(file_name.c_str(), &bmpClsid);

    return bmp;
 });

 // Deletes Bitmap objects.
 transformer<Bitmap*, Bitmap*> delete_bitmap([](Bitmap* bmp) -> Bitmap* {      
    delete bmp;
    return nullptr;
 });

 // Decrements the event counter.
 call<Bitmap*> decrement([&](Bitmap* _) {      
    active.signal();
 });

Add the following code to connect the network.

//
// Connect the network.
//   

load_bitmap.link_target(&loaded_bitmaps);

loaded_bitmaps.link_target(&grayscale);
loaded_bitmaps.link_target(&colormask);   
colormask.link_target(&darken);
loaded_bitmaps.link_target(&sepiatone);
loaded_bitmaps.link_target(&decrement);

grayscale.link_target(&save_bitmap);
darken.link_target(&save_bitmap);
sepiatone.link_target(&save_bitmap);

save_bitmap.link_target(&delete_bitmap);
delete_bitmap.link_target(&decrement);

Add the following code to send to the head of the network the full path of each JPEG file in the directory.

// Traverse all files in the directory.
wstring searchPattern = directory;
searchPattern.append(L"\\*");

WIN32_FIND_DATA fileFindData;
HANDLE hFind = FindFirstFile(searchPattern.c_str(), &fileFindData);
if (hFind == INVALID_HANDLE_VALUE) 
   return;
do
{
   if (!(fileFindData.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY))
   {
      wstring file = fileFindData.cFileName;

      // Process only JPEG files.
      if (file.rfind(L".jpg") == file.length() - 4)
      {
         // Form the full path to the file.
         wstring full_path(directory);
         full_path.append(L"\\");
         full_path.append(file);

         // Increment the count of work items.
         active.add_count();

         // Send the path name to the network.
         send(load_bitmap, full_path);
      }
   }
}
while (FindNextFile(hFind, &fileFindData) != 0); 
FindClose(hFind);

Wait for the countdown_event variable to reach zero.

// Wait for all operations to finish.
active.wait();

The following table describes the members of the network.

Member	Description
`load_bitmap`	A concurrency::transformer object that loads a `Bitmap` object from disk and adds an entry to the `map` object to associate the image with its original file name.
`loaded_bitmaps`	A concurrency::unbounded_buffer object that sends the loaded images to the image processing filters.
`grayscale`	A `transformer` object that converts images that are authored by Tom to grayscale. It uses the metadata of the image to determine its author.
`colormask`	A `transformer` object that removes the green and blue color components from images that have red as the dominant color.
`darken`	A `transformer` object that darkens images that have red as the dominant color.
`sepiatone`	A `transformer` object that applies sepia toning to images that are not authored by Tom and are not predominantly red.
`save_bitmap`	A `transformer` object that saves the processed `image` to disk as a bitmap. `save_bitmap` retrieves the original file name from the `map` object and changes its file name extension to .bmp.
`delete_bitmap`	A `transformer` object that frees the memory for the images.
`decrement`	A concurrency::call object that acts as the terminal node in the network. It decrements the `countdown_event` object to signal to the main application that an image has been processed.

The loaded_bitmaps message buffer is important because, as an unbounded_buffer object, it offers Bitmap objects to multiple receivers. When a target block accepts a Bitmap object, the unbounded_buffer object does not offer that Bitmap object to any other targets. Therefore, the order in which you link objects to an unbounded_buffer object is important. The grayscale, colormask, and sepiatone message blocks each use a filter to accept only certain Bitmap objects. The decrement message buffer is an important target of the loaded_bitmaps message buffer because it accepts all Bitmap objects that are rejected by the other message buffers. An unbounded_buffer object is required to propagate messages in order. Therefore, an unbounded_buffer object blocks until a new target block is linked to it and accepts the message if no current target block accepts that message.

If your application requires that multiple message blocks process the message, instead of just the one message block that first accepts the message, you can use another message block type, such as overwrite_buffer. The overwrite_buffer class holds one message at a time, but it propagates that message to each of its targets.

The following illustration shows the image processing network:

Image processing network.

The countdown_event object in this example enables the image processing network to inform the main application when all images have been processed. The countdown_event class uses a concurrency::event object to signal when a counter value reaches zero. The main application increments the counter every time that it sends a file name to the network. The terminal node of the network decrements the counter after each image has been processed. After the main application traverses the specified directory, it waits for the countdown_event object to signal that its counter has reached zero.

The following example shows the countdown_event class:

// A synchronization primitive that is signaled when its 
// count reaches zero.
class countdown_event
{
public:
   countdown_event(unsigned int count = 0)
      : _current(static_cast<long>(count)) 
   {
      // Set the event if the initial count is zero.
      if (_current == 0L)
         _event.set();
   }
     
   // Decrements the event counter.
   void signal() {
      if(InterlockedDecrement(&_current) == 0L) {
         _event.set();
      }
   }

   // Increments the event counter.
   void add_count() {
      if(InterlockedIncrement(&_current) == 1L) {
         _event.reset();
      }
   }
   
   // Blocks the current context until the event is set.
   void wait() {
      _event.wait();
   }
 
private:
   // The current count.
   volatile long _current;
   // The event that is set when the counter reaches zero.
   event _event;

   // Disable copy constructor.
   countdown_event(const countdown_event&);
   // Disable assignment.
   countdown_event const & operator=(countdown_event const&);
};

[Top]

The Complete Example

The following code shows the complete example. The wmain function manages the GDI+ library and calls the ProcessImages function to process the JPEG files in the Sample Pictures directory.

// image-processing-network.cpp
// compile with: /DUNICODE /EHsc image-processing-network.cpp /link gdiplus.lib
#include <windows.h>
#include <gdiplus.h>
#include <iostream>
#include <map>
#include <agents.h>
#include <ppl.h>

using namespace concurrency;
using namespace Gdiplus;
using namespace std;

// Retrieves the red, green, and blue components from the given
// color value.
void GetRGB(DWORD color, BYTE& r, BYTE& g, BYTE& b)
{
   r = static_cast<BYTE>((color & 0x00ff0000) >> 16);
   g = static_cast<BYTE>((color & 0x0000ff00) >> 8);
   b = static_cast<BYTE>((color & 0x000000ff));
}

// Creates a single color value from the provided red, green, 
// and blue components.
DWORD MakeColor(BYTE r, BYTE g, BYTE b)
{
   return (r<<16) | (g<<8) | (b);
}

// Calls the provided function for each pixel in a Bitmap object.
void ProcessImage(Bitmap* bmp, const function<void (DWORD&)>& f)
{
   int width = bmp->GetWidth();
   int height = bmp->GetHeight();

   // Lock the bitmap.
   BitmapData bitmapData;
   Rect rect(0, 0, bmp->GetWidth(), bmp->GetHeight());
   bmp->LockBits(&rect, ImageLockModeWrite, PixelFormat32bppRGB, &bitmapData);

   // Get a pointer to the bitmap data.
   DWORD* image_bits = (DWORD*)bitmapData.Scan0;

   // Call the function for each pixel in the image.
   parallel_for (0, height, [&, width](int y)
   {      
      for (int x = 0; x < width; ++x)
      {
         // Get the current pixel value.
         DWORD* curr_pixel = image_bits + (y * width) + x;

         // Call the function.
         f(*curr_pixel);
      }
   });

   // Unlock the bitmap.
   bmp->UnlockBits(&bitmapData);
}

// Converts the given image to grayscale.
Bitmap* Grayscale(Bitmap* bmp) 
{
   ProcessImage(bmp, 
      [](DWORD& color) {
         BYTE r, g, b;
         GetRGB(color, r, g, b);

         // Set each color component to the average of 
         // the original components.
         BYTE c = (static_cast<WORD>(r) + g + b) / 3;
         color = MakeColor(c, c, c);
      }
   );
   return bmp;
}

// Applies sepia toning to the provided image.
Bitmap* Sepiatone(Bitmap* bmp) 
{
   ProcessImage(bmp, 
      [](DWORD& color) {
         BYTE r0, g0, b0;
         GetRGB(color, r0, g0, b0);

         WORD r1 = static_cast<WORD>((r0 * .393) + (g0 *.769) + (b0 * .189));
         WORD g1 = static_cast<WORD>((r0 * .349) + (g0 *.686) + (b0 * .168));
         WORD b1 = static_cast<WORD>((r0 * .272) + (g0 *.534) + (b0 * .131));

         color = MakeColor(min(0xff, r1), min(0xff, g1), min(0xff, b1));
      }
   );
   return bmp;
}

// Applies the given color mask to each pixel in the provided image.
Bitmap* ColorMask(Bitmap* bmp, DWORD mask)
{
   ProcessImage(bmp, 
      [mask](DWORD& color) {
         color = color & mask;
      }
   );
   return bmp;
}

// Darkens the provided image by the given amount.
Bitmap* Darken(Bitmap* bmp, unsigned int percent)
{
   if (percent > 100)
      throw invalid_argument("Darken: percent must less than 100.");

   double factor = percent / 100.0;

   ProcessImage(bmp, 
      [factor](DWORD& color) {
         BYTE r, g, b;
         GetRGB(color, r, g, b);
         r = static_cast<BYTE>(factor*r);
         g = static_cast<BYTE>(factor*g);
         b = static_cast<BYTE>(factor*b);
         color = MakeColor(r, g, b);
      }
   );
   return bmp;
}

// Determines which color component (red, green, or blue) is most dominant
// in the given image and returns a corresponding color mask.
DWORD GetColorDominance(Bitmap* bmp)
{
   // The ProcessImage function processes the image in parallel.
   // The following combinable objects enable the callback function
   // to increment the color counts without using a lock.
   combinable<unsigned int> reds;
   combinable<unsigned int> greens;
   combinable<unsigned int> blues;

   ProcessImage(bmp, 
      [&](DWORD& color) {
         BYTE r, g, b;
         GetRGB(color, r, g, b);
         if (r >= g && r >= b)
            reds.local()++;
         else if (g >= r && g >= b)
            greens.local()++;
         else
            blues.local()++;
      }
   );
   
   // Determine which color is dominant and return the corresponding
   // color mask.

   unsigned int r = reds.combine(plus<unsigned int>());
   unsigned int g = greens.combine(plus<unsigned int>());
   unsigned int b = blues.combine(plus<unsigned int>());

   if (r + r >= g + b)
      return 0x00ff0000;
   else if (g + g >= r + b)
      return 0x0000ff00;
   else
      return 0x000000ff;
}

// Retrieves the class identifier for the given MIME type of an encoder.
int GetEncoderClsid(const WCHAR* format, CLSID* pClsid)
{
   UINT  num = 0;          // number of image encoders
   UINT  size = 0;         // size of the image encoder array in bytes

   ImageCodecInfo* pImageCodecInfo = nullptr;

   GetImageEncodersSize(&num, &size);
   if(size == 0)
      return -1;  // Failure

   pImageCodecInfo = (ImageCodecInfo*)(malloc(size));
   if(pImageCodecInfo == nullptr)
      return -1;  // Failure

   GetImageEncoders(num, size, pImageCodecInfo);

   for(UINT j = 0; j < num; ++j)
   {
      if( wcscmp(pImageCodecInfo[j].MimeType, format) == 0 )
      {
         *pClsid = pImageCodecInfo[j].Clsid;
         free(pImageCodecInfo);
         return j;  // Success
      }    
   }

   free(pImageCodecInfo);
   return -1;  // Failure
}

// A synchronization primitive that is signaled when its 
// count reaches zero.
class countdown_event
{
public:
   countdown_event(unsigned int count = 0)
      : _current(static_cast<long>(count)) 
   {
      // Set the event if the initial count is zero.
      if (_current == 0L)
         _event.set();
   }
     
   // Decrements the event counter.
   void signal() {
      if(InterlockedDecrement(&_current) == 0L) {
         _event.set();
      }
   }

   // Increments the event counter.
   void add_count() {
      if(InterlockedIncrement(&_current) == 1L) {
         _event.reset();
      }
   }
   
   // Blocks the current context until the event is set.
   void wait() {
      _event.wait();
   }
 
private:
   // The current count.
   volatile long _current;
   // The event that is set when the counter reaches zero.
   event _event;

   // Disable copy constructor.
   countdown_event(const countdown_event&);
   // Disable assignment.
   countdown_event const & operator=(countdown_event const&);
};

// Demonstrates how to set up a message network that performs a series of 
// image processing operations on each JPEG image in the given directory and
// saves each altered image as a Windows bitmap.
void ProcessImages(const wstring& directory)
{
   // Holds the number of active image processing operations and 
   // signals to the main thread that processing is complete.
   countdown_event active(0);

   // Maps Bitmap objects to their original file names.
   map<Bitmap*, wstring> bitmap_file_names;
      
   //
   // Create the nodes of the network.
   //

   // Loads Bitmap objects from disk.
   transformer<wstring, Bitmap*> load_bitmap(
      [&](wstring file_name) -> Bitmap* {
         Bitmap* bmp = new Bitmap(file_name.c_str());
         if (bmp != nullptr)
            bitmap_file_names.insert(make_pair(bmp, file_name));
         return bmp;
      }
   );

   // Holds loaded Bitmap objects.
   unbounded_buffer<Bitmap*> loaded_bitmaps;
  
   // Converts images that are authored by Tom to grayscale.
   transformer<Bitmap*, Bitmap*> grayscale(
      [](Bitmap* bmp) {
         return Grayscale(bmp);
      },
      nullptr,
      [](Bitmap* bmp) -> bool {
         if (bmp == nullptr)
            return false;

         // Retrieve the artist name from metadata.
         UINT size = bmp->GetPropertyItemSize(PropertyTagArtist);
         if (size == 0)
            // Image does not have the Artist property.
            return false;

         PropertyItem* artistProperty = (PropertyItem*) malloc(size);
         bmp->GetPropertyItem(PropertyTagArtist, size, artistProperty);
         string artist(reinterpret_cast<char*>(artistProperty->value));
         free(artistProperty);
         
         return (artist.find("Tom ") == 0);
      }
   );
   
   // Removes the green and blue color components from images that have red as
   // their dominant color.
   transformer<Bitmap*, Bitmap*> colormask(
      [](Bitmap* bmp) {
         return ColorMask(bmp, 0x00ff0000);
      },
      nullptr,
      [](Bitmap* bmp) -> bool { 
         if (bmp == nullptr)
            return false;
         return (GetColorDominance(bmp) == 0x00ff0000);
      }
   );

   // Darkens the color of the provided Bitmap object.
   transformer<Bitmap*, Bitmap*> darken([](Bitmap* bmp) {
      return Darken(bmp, 50);
   });

   // Applies sepia toning to the remaining images.
   transformer<Bitmap*, Bitmap*> sepiatone(
      [](Bitmap* bmp) {
         return Sepiatone(bmp);
      },
      nullptr,
      [](Bitmap* bmp) -> bool { return bmp != nullptr; }
   );

   // Saves Bitmap objects to disk.
   transformer<Bitmap*, Bitmap*> save_bitmap([&](Bitmap* bmp) -> Bitmap* {
      // Replace the file extension with .bmp.
      wstring file_name = bitmap_file_names[bmp];
      file_name.replace(file_name.rfind(L'.') + 1, 3, L"bmp");
      
      // Save the processed image.
      CLSID bmpClsid;
      GetEncoderClsid(L"image/bmp", &bmpClsid);      
      bmp->Save(file_name.c_str(), &bmpClsid);

      return bmp;
   });

   // Deletes Bitmap objects.
   transformer<Bitmap*, Bitmap*> delete_bitmap([](Bitmap* bmp) -> Bitmap* {      
      delete bmp;
      return nullptr;
   });

   // Decrements the event counter.
   call<Bitmap*> decrement([&](Bitmap* _) {      
      active.signal();
   });

   //
   // Connect the network.
   //   
   
   load_bitmap.link_target(&loaded_bitmaps);
   
   loaded_bitmaps.link_target(&grayscale);
   loaded_bitmaps.link_target(&colormask);   
   colormask.link_target(&darken);
   loaded_bitmaps.link_target(&sepiatone);
   loaded_bitmaps.link_target(&decrement);
   
   grayscale.link_target(&save_bitmap);
   darken.link_target(&save_bitmap);
   sepiatone.link_target(&save_bitmap);
   
   save_bitmap.link_target(&delete_bitmap);
   delete_bitmap.link_target(&decrement);
   
   // Traverse all files in the directory.
   wstring searchPattern = directory;
   searchPattern.append(L"\\*");

   WIN32_FIND_DATA fileFindData;
   HANDLE hFind = FindFirstFile(searchPattern.c_str(), &fileFindData);
   if (hFind == INVALID_HANDLE_VALUE) 
      return;
   do
   {
      if (!(fileFindData.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY))
      {
         wstring file = fileFindData.cFileName;

         // Process only JPEG files.
         if (file.rfind(L".jpg") == file.length() - 4)
         {
            // Form the full path to the file.
            wstring full_path(directory);
            full_path.append(L"\\");
            full_path.append(file);

            // Increment the count of work items.
            active.add_count();

            // Send the path name to the network.
            send(load_bitmap, full_path);
         }
      }
   }
   while (FindNextFile(hFind, &fileFindData) != 0); 
   FindClose(hFind);
      
   // Wait for all operations to finish.
   active.wait();
}

int wmain()
{
   GdiplusStartupInput gdiplusStartupInput;
   ULONG_PTR           gdiplusToken;

   // Initialize GDI+.
   GdiplusStartup(&gdiplusToken, &gdiplusStartupInput, nullptr);

   // Perform image processing.
   // TODO: Change this path if necessary.
   ProcessImages(L"C:\\Users\\Public\\Pictures\\Sample Pictures");

   // Shutdown GDI+.
   GdiplusShutdown(gdiplusToken);
}

The following illustration shows sample output. Each source image is above its corresponding modified image.

Sample output for the example.

Lighthouse is authored by Tom Alphin and therefore is converted to grayscale. Chrysanthemum, Desert, Koala, and Tulips have red as the dominant color and therefore have the blue and green color components removed and are darkened. Hydrangeas, Jellyfish, and Penguins match the default criteria and therefore are sepia toned.

[Top]

Compiling the Code

Copy the example code and paste it in a Visual Studio project, or paste it in a file that is named image-processing-network.cpp and then run the following command in a Visual Studio Command Prompt window.

cl.exe /DUNICODE /EHsc image-processing-network.cpp /link gdiplus.lib

Share via

Walkthrough: Creating an Image-Processing Network

Prerequisites

Sections

Defining Image Processing Functionality

Creating the Image Processing Network

To create the image processing network

The Complete Example

Compiling the Code

See also

Feedback

Additional resources