14.3 Video processing part 2: grayscale and color  (Page 3/3)

Code briefing

The code provided merely performs a color conversion required to go from the input NTSC image to the output RGB image. The relevant streams of data are brought in using the in_luma, in_cr, in_cb odd and even streams

The odd, even is done because the input YcbCr data is interlaced and the different "color" components Y(luminance), Cr, and Cb are stored in different arrays, unlike RGB where the data is packed together for each pixel. Thus the streams are accessed inside the color_conv_image wrapper function. We then pass a line at a time to the color_conv component function which converts and flips one line at a time.

We will need to modify the code here, in color_conv to achieve your goals. The control surface will be a square block 100 by 100 pixels in the bottom left corner of the screen. The brightness will be calculated by summing all the R, G and B values of all the pixels in this portion of the screen. We then apply the tint effect as such:

• if the total brightness is below a certain level 'X': use a red tint,
• if the total brightness is above 'X' and below 'Y' : use a green tint,
• if above 'Y' : use a blue tint

The tint has to be scaled too. For example, if brightness is less than X but close to it we need a high blue. But if it's closer to zero we need a darker blue and so on. The scaling need not be linear. In fact if you did the auto-contrast function you will have noticed that the floating point operations are expensive, they tend to slow the system. This is more so in the color case, as we have more data to scale. So try to use simple bit shifts to achieve the needed effect.

• Right Shift : >>
• Left Shift : <<
• Masking : Use a single ampersand, so to extract the first red component: RGB[1] & 0xF8

Tips and tricks

You're on your own now! But some things to remember and to watch out for are presented here, as well as ideas for improvement. Remember:

• The input is two bytes per pixel. Keep the packed RGB format in mind.
• Also we process one line at a time from top to bottom. We cannot go back to previous lines to change them. So we can only modify the tint of the screen below the control surface. What you could do however is keep global variables for the different scalings in main. Then pass these to color_conv by reference, and update it when converting colors. But perform the update after using the existing scale values to scale the screen region above the control surface. This will introduce a delay from scaling change to screen update. This can be solved by copying the entire input to memory before outputting it but this is quite expensive, and we'll deal with memory in the next section.
• Be careful when performing masking, shifting and separting. Bring things down to least significant set of bits (within a byte) to simplify thinking of the scaling. Also be careful not to overlap masks, especially during shifting and adding

Here are a few recommendations:

• Try to use the Y data passed to the color_con funtion to compute the brightness – much faster.
• Also poke around and find out how to use the Cr, Cb data and scale those. It's far less expensive and may produce neater results.
• If something doesn't work, think things through again. Or better still take a break and then come back to the problem.