Smart resize for monochrome bitmaps

If we make a monochrome bitmap smaller by a simple pixel resampling the resulting image looks lousy (pic.2). The problem here is that we take a stand-alone pixel and disregard its surrounding. We can improve the picture a little bit by calculating the value of a pixel as the average of the pixels we disregard and then rounding this value to the nearest color. However the best result is obtained in a grayscale destination bitmap (pic. 3). This is implemented in the following function.
 
Pic.1 Original image Pic 2. Simple resize   Pic 3. Smart resize

#define WIDTHBYTES(bits)    (((bits) + 31) / 32 * 4) // for padding

///////////////////////////////////////////////////////////////////
// Function name    : ZoomOutBmp
// Description      : creates a new bitmap which is a grayscale 
//                    zoomed out version of the original
// Return type      : HDIB - handle to a new bitmap 
// Argument         : double zoom - number of times to zoom out
// Argument         : HDIB hSrcDIB - handle to a source bitmap
///////////////////////////////////////////////////////////////////
HDIB ZoomOutBmp(double zoom, HDIB hSrcDIB)
{
    if (hSrcDIB == NULL) // nothing to do
        return NULL;
    
    if (zoom < 1) // no zoomin in this function
        return NULL;
    
    LPSTR pSrcDIB = (LPSTR) ::GlobalLock((HGLOBAL) hSrcDIB);
    
    BITMAPINFOHEADER& bmihSrc = *(BITMAPINFOHEADER*)pSrcDIB;
    ASSERT(bmihSrc.biBitCount == 1); // only monochrome bitmaps supported
    LPSTR pSrcBits = (LPSTR) (pSrcDIB + sizeof(BITMAPINFOHEADER) + sizeof(RGBQUAD)*2);
    
    BITMAPINFOHEADER bmihDst = bmihSrc;
    bmihDst.biWidth = (LONG)(bmihDst.biWidth / zoom + 0.5);
    bmihDst.biHeight = (LONG)(bmihDst.biHeight / zoom + 0.5);
    bmihDst.biBitCount = 8; // grayscale in any case
    bmihDst.biClrUsed = 0;
    
    // prepare destination bitmap
    DWORD dwDIBSize = sizeof(bmihDst) + sizeof(RGBQUAD)*256 +
        WIDTHBYTES(bmihDst.biWidth * bmihDst.biBitCount) * bmihDst.biHeight;
    
    bmihDst.biSizeImage = dwDIBSize;
    
    // allocate space for the new bitmap
    HDIB hDstDIB = (HDIB) ::GlobalAlloc(GMEM_MOVEABLE | GMEM_ZEROINIT, dwDIBSize);
    
    if (hDstDIB == 0) {
        ::GlobalUnlock((HGLOBAL) hSrcDIB);
        return NULL;
    }
    
    LPSTR pDstDIB = (LPSTR) ::GlobalLock((HGLOBAL) hDstDIB);
    
    // copy header
    memcpy(pDstDIB, &bmihDst, sizeof(bmihDst));
    
    // prepare grayscale palette
    for (int i=0; i < (1 << bmihDst.biBitCount); i++) {
        
        RGBQUAD& palEntry = *(RGBQUAD*)(pDstDIB + sizeof(bmihDst) + i * sizeof(RGBQUAD));
        palEntry.rgbRed = palEntry.rgbGreen = palEntry.rgbBlue = i;
        
    }
    
    LPSTR pDstBits = (LPSTR) (pDstDIB + sizeof(BITMAPINFOHEADER) + sizeof(RGBQUAD)*256);
    
    // now fill the bits
    LPSTR curSrcLineBits, curDstLineBits;
    int j, k;
    int scale = (int)(zoom + 0.5); // integer zoom out factor, i.e. 1:5
    int hBase, vBase;
    unsigned char value;
    
    // for every _scale_ lines in a source bitmap we will get one line
    // in the destination bitmap. Similarly for _scale_ columns in the
    // source we'll obtain one destination column.
    
    for (int strip=0; strip < bmihDst.biHeight; strip++) { // for every dst line
        
        curDstLineBits = pDstBits + strip * WIDTHBYTES(bmihDst.biWidth * bmihDst.biBitCount);
        vBase = int(strip * zoom + 0.5);
        
        for (i=0; i < scale; i++) {  // accumulate _scale_ rows
            
            curSrcLineBits = pSrcBits + (vBase + i) * WIDTHBYTES(bmihSrc.biWidth * bmihSrc.biBitCount);
            
            // prepare horizontally condensed lines for this strip
            for (j=0; j < bmihDst.biWidth; j++) { // for all bits in line
                
                hBase = int(j * zoom + 0.5); // mapped index on source
                for (k=0; k < scale; k++) { // accumulate _scale_ columns
                    
                    value = (curSrcLineBits[(hBase+k)/8] & (1 << (7 - (hBase+k)%8))) ? 0xff : 0;
                    curDstLineBits[j] += value / scale / scale; // main accumulator
                }
            }
        }
        
    }
    
    // unlock memory
    ::GlobalUnlock((HGLOBAL) hSrcDIB);
    ::GlobalUnlock((HGLOBAL) hDstDIB);
    
    return hDstDIB;
}

This function may be improved in several ways, for example, working with a weighted fraction of a pixel or centering the accumulated source pixels. However, the resulting visual impovement is not significant.

Download demo project - 38 KB



Comments

Leave a Comment
  • Your email address will not be published. All fields are required.

Top White Papers and Webcasts

  • At present, the most commonly deployed parallel file system is Lustre, and its adoption is growing across the HPC industry. According to Intersect 360 Research, "Through its leadership and involvement with OpenSFS, Cray advocates for the development of features that drive efficient performance at scale." Moreover, with help from Cray and OpenSFS, Lustre is gaining greater adoption across broader commercial application categories. As data and I/O requirements grow in commercial markets, technology vendors -- …

  • As virtualization becomes the norm throughout organizations of nearly all sizes, and as more organizations look to private cloud solutions, IT decision makers are increasingly in need of ways to keep storage costs and complexity under control in the face of often-runaway virtual machine (VM) sprawl. Application-aware storage is designed to help achieve these important goals. Read this white paper to learn how application-aware storage allows you to gain VM-level visibility into application performance and …

Most Popular Programming Stories

More for Developers

RSS Feeds

Thanks for your registration, follow us on our social networks to keep up-to-date