THE MAGAZINE FOR COMPUTER APPLICATIONS
Circuit Cellar Online offers articles illustrating creative solutions
and unique applications through complete projects, practical
tutorials, and useful design techniques.

LOOK MA, NO PC!

A $55 Webcam
by Steve Freyder, David Helland, & Bruce LIghtner

Start • Hardware Design • Picking a Camera • Cheap CMOS Cameras • PicoWeb Server Hardware • Firmware Functions • Software Design • PicoWeb Firmware • Java Applet • Smile • Source and PDF

Java Applet

A Java applet was necessary for this project because the image data returned from the Barbie Photo Designer camera needs to be processed before it can be displayed. The camera does not store images in a format that can be directly displayed by a web browser (e.g., JPEG or GIF images). Instead, the camera sends raw image data to the computer in the form of a Bayer color pattern. The Java code causes a TCP/IP socket to be opened by the web browserıs computer to transfer the raw picture data from the PicoWeb server, and then to process the data as necessary into a viewable image.

Figure 5ıThis is the pattern of colored filters that covers the image sensor chipıs 160 ı 120 array of light sensors (pixels). The "missing" red, green, and blue pixel values must be interpolated from neighboring pixels of like color.

The raw pixels from the camera come from a 2 ı 2 red-green-blue-green Bayer array as shown in Figure 5. Each pixel in the image sensor chip is covered by a colored filter according to the Bayer pattern shown. There are two green pixels for every red and every blue pixel. We need to supply a red, green, and blue pixel for every possible pixel location in order to derive a real image. If we donıt do this, we get a low-resolution greenish image, as shown in Photo 3a. We do this by looking at like-colored pixels in the neighborhood and making an intelligent guess about the probable color and intensity of the light that struck each pixel when the photo was snapped. (Next time you read about the latest full-color digital camera with 2.1 million pixels, remember that in some sense, two-thirds of the pixel data is made up!)

a)

b)

c)

Photo 3ıThe first image (a) shows the raw pixel data from the camera (Bayer color pattern). This image is processed by the PicoWebCam-supplied Java applet to supply the "missing" pixels (b), then it is sharpened (c). Believe it or not, when not enlarged, the sharpened image looks better to most people.

We have lots of pixel interpolation algorithms to choose from, of varying complexity, and with a wide range of computational requirements. Our Java applet executes a simple, fast nearest-neighbor bilinear interpolation algorithm to quickly provide a full-color image from the raw pixel array. The resulting image is then sharpened using a convolution function before display. This is something that Mattelıs PC software does in order to make their cameraıs otherwise tiny fuzzy photos look better. Photo 3b and 3c show an enlargement of a sample camera image after processing by the PicoWebCamıs Java applet.

An excellent discussion of Bayer color pattern processing algorithms is titled "A Study of Spatial Color Interpolation Algorithms for Single-Detector Digital Cameras", by Ting Chen. [1] The basic algorithm for the image sharpening was inspired by an article titled "Image Processing with Java 2D", by Bill Day and Jonathan Knudsen. [2] The Java compiler used for this project was provided by Sun Microsystems. A complete, free Java program development kit (JAVATM 2 SDK, Standard Edition Version 1.3) is available for downloading from Sun.

PREVIOUS • NEXT