[The J Maker Home] [Program Home] [Pricing] [Contact] [FAQ]
	Lab with Image Utilities (LIU) Examples
The Package:
Introduction: The entire LIU package contains 90+ image programs with 24300+ lines of code. Most of the ideas in LIU are inspired from general image processing, computer vision and graphics books, and from commercial packages like PhotoShop, PhotoVision, XV, Image Magick, GIMP, etc. Some of the algorithms are my own inventions, especially in the category, "Special Effect". Categories include boundary detection (BD), contrast enhancement (CE), fundamental tools (FT), geometric transformations (GT), image segmentation (IS), mensuration (M), noise simulation / reduction (NSR), and special effect (SE). Programs can be run either individually or invoked from the main menu. They can start as applications or applets. While executed as an applet, it will not be able to save the results due to Java's security reasons and the subprogram "data" will not function properly. Reading multiple local and URL images is supported. Award: This applet was the second place cash award winner of the ACM / IBM 1999 Quest for Java Programming Contest. Click here to read the interview at ACM's Contest site Download: [program download] [examples download] File Description: liu.html - staring HTML liutm.gif - LIU logo LIU*.class - library classes liu.class - starting application liuapplet.class - staring applet liu*.class - used by main menu "liu" GIF*.class - GIF encoder other *.class - individual programs liu*.gif - LIU program images x*.gif x*.jpg - test images proto.java - sample source code to use the image library Author: The J Maker (webmaster@thejmaker.com)
Quick Execution Guide:
General Steps: For example, if you want to run the "ripple" program, you can do any of the followings: Run as an application by starting the main menu from the command line with "java liu" Select "ripple" from "Programs" menu Enter "x1.gif" in input file list Enter "y1.gif" in output file list Click on "run" button Run as an application by starting "ripple" from the command line with "java ripple x1.gif y1.gif" Run as an applet by starting the main menu from the command line with "appletviewer liu.html" Run as an applet by starting the main menu (liu.html) through the LIU launcher in a web browser (again, it will not save your result and subprogram "data" will not run properly) Mouse Input: In addition to general mouse actions in JDK, some mouse input types on an image are implemented in LIU. They include: click to find out the current position in an image (Fig. 1), e.g. "affine" and most subprograms drag to get the frame containing the region (Fig. 2), e.g. "blossom" and most subprograms click to match positions between a pair of images to manually establish correspondences (Fig. 3), e.g. "morph" drag to get the vector representing point movement (Fig. 4), e.g. "warp" Fig. 1 Fig. 2 Fig. 3 Fig. 4 Text Input: A few points are to be noted here: Properly modify all default values if desired. Some programs can take a variable number of filenames, e.g. "array", "view", etc. Supply your input filenames with URLs when necessary, e.g. http://www.nowhere.com/test.gif
Technical Background:
As mentioned at the beginning of this documentation, this package utilizes knowledge from computer vision, image processing, computer graphics, and some ideas from commercial packages. A few programs are started from scratch and thus original. I would rather not go into mathematical details here, e.g. matrix algebra or Fourier analysis because that will probably take up lengthy pages just to make people understand the magic behind these programs. Some references are attached at the end for interested readers. There are some major techniques explained as follows: Java-related: Some missing GUI components in JDK 1.1 are implemented just to make the interface more desirable, e.g. image button, labeled panel, etc. Immediate garbage collection is done to keep the memory usage more managable. Internal text parsing is done to handle variable number of file processing. A general purpose image access library (starting with LIU*) is implemented and used in all subprograms. The package is designed to run in different modes - independent or comprehensive, applet or application. Global Spline Warping: In morph and warp, a global spline is defined by a set of vector bases computed from user's mouse inputs. With the basis function, all the points are transformed (warped) into new locations. This requires solving the matrix inverse by LU decomposition. Cross Dissolve: In order to smoothly transform from image A to B, creating transition frames in between needs to take two things into account - position and color. This effect can be achieved by a combination of warping A to B and warping B to A simultaneously. Cross dissolve is thus to create smooth transitions. Fig. 5 and Fig. 6 are our image A and B. Fig. 7 is the 31st frame out of 60 in the sequence. Notice in Fig. 7 that the positions and colors of the features (e.g. eyes) have changed. Fig. 5 Fig. 6 Fig. 7 Double Buffering: The idea is actually very simple. We first prepare an offscreen duplicate graphics object. Then draw everything on this object. After all the drawing work has been accomplished, we then replace the onscreen image with this offscreen object. This is used to reduce the flickering, especially for animation. Bilinear Interpolation: For example, when we zoom in an image, lots of pixels holes need to be filled in to see a bigger image. By simply using the color of the closest pixel (called nearest neighbor), we will sometimes get undesirable zigzag artifacts. (called aliasing) Therefore, interpolating the color with four closest corners is used here for better results. This is an area-based approach. Inverse Mapping: When transforming a pixel to a new location, instead of applying the transformation matrix to the pixel and then finding out the new loation, it is done backwards in the package. That is - a new location is first assumed and then its orignal location is computed. Thus it is called inverse mapping. This will make the computation more efficient in bilinear interpolation. Physically-Based Deformable Model: As in scissors, this idea is to put a rough initial boundary close to the desired object and hopefully this boundary will automatically deform to "exactly" overlap with the object boundary. Since it uses material properties (stretching and bending), it is thus called. This involves in solving a nonlinear differential equation.
A Complete Listing of Examples:
Since this image package includes quite a number of utilities and has grown somewhat out of proportion, sometimes it is even difficult for myself to choose desirable utility programs in such a collection. These LIU examples serve in two ways: one is to give you a quick overview of what LIU can do, while the other is to act as a learn-from-example manual. Programs with are more emphasized in terms of complexity and originality. For example, morph is one of the most complex programs in the package because techniques in solving a system of linear equations and interpolating with cross dissolve are used. affine performs general affine transformations on the image, i.e. translation, scaling, rotation, shear array makes an image array bblur performs binomial blurring in a selected area bitplane outputs a selected bitplane on three color channels blossom blossoms the image blur blurs the image by averaging neighboring pixels bright adjusts brightness button creates a 3-D button cblur performs circular blurring on the image cmap displays a color map color converts the image colors into specified range colorpix only colors pixels of special interest compose composes two images with bitwise operations concave emulates the concave mirror effect contrast adjusts the image contrast by pixel mean convex emulates the convex mirror effect crop crops the image in a selected region cube maps the image onto a cube data displays the image data in hexadecimal and ASCII modes disarray disarrays an image dither performs dither halftoning dizzy emulates the dizzy effect drain computes the drainage pattern of the image on three color channels dwmtm performs double window modified trimmed mean filtering with specified windows emboss emulates the emboss effect expand adds some space to an image extrema finds local extrema flip flips the image along X, Y, or diagonal axis focus focuses on a portion of the image frame adds a frame to the image glass emulates the rectangular glass effect gnoise adds additive Gaussian noise to the image with specified mean and variance gpen emulates the graphic pen effect gradient computes the image gradient magnitude grey converts the image into grey with specified range half duplicates half an image height visualizes the image height in 3-D highlite highlights a selected region histeq performs histogram equalization impress emulates the impressionistic effect inv inverts the image color laplacian computes Laplacian transform layer emulates the layer effect with specified transformations level draws the level curve light emulates a light source line detects lines maxmin performs maxmin filtering with a sized window median performs median filtering with a sized window memo makes a photo memo mesh generates the image mesh mmse performs adaptive minimum mean square error filtering with a sized window and specified noise variance morph morphs one image into another (concatenated by GIF Construction Set) mosaic emulates the mosaic effect in a selected region motion performs motion blur nenoise generates negative exponential noise to the image with specified variance (for laser speckle) newspaper performs newspaper halftoning oil emulates the oil painting effect pastel emulates the pastel effect pnoise adds Poisson noise to the image with specified mean and standard deviation point detects points poster posterizes the image proto serves as a programming template pull pulls an image region push pushes an image region raise raises blocks in the image randomly redim changes image dimensions rgb outputs R, G, B channels of the image ripple emulates the ripple effect of the image at a selected center saturate saturates colors scissors cuts and pastes a portion of the first image scramble emulates the scramble effect shade performs simple Lambertian shading shadow creates an image shadow sharpen sharpens the image sketch emulates the sketch effect sphere maps an image onto a sphere spnoise adds salt and pepper noise to the image with specified probability spread spreads intensities stretch stretches the image at four corers text adds text onto the image stretch creates a text mask thermo emulates the thermo effect of the image thresh threshholds the image into black and white tile tiles the image tint adjusts the color tint angle triangle triangularizes the image twirl emulates the twirl effect unoise generates additive uniform noise to the image with specified mean and variance vibrate emulates the vibration effect view can display multiple images like a mini VCR warp warps the image according to mouse inputs wave emulates the visual effect of the image sitting in front of water wind emulates the wind effect wrinkle emulates the wrinkle effect
References:
Ken Arnold and James Gosling, "The Java Programming Language", Addison-Wesley, 1997 Henning Bassmann and Philipp W. Besslich, "Ad Oculos Digital Image Processing", International Thomson Publishing, 1995 Mary Campione and Kathy Walrath, "The Java Tutorial", Addison-Wesley, 1997 M. Cohn, B. Morgan, M. Morrison, M. Nygard, D. Joshui, and T. Trinko, "Java Developer's Reference", Sams.net Publishing, 1996 Foley, van Dam, Feiner, and Hughes, "Computer Graphics", Addison-Wesley Publishing, 1990 Rafael C. Gonzalez and Richard E. Woods, "Digital Image Processing", Addison-Wesley Publishing, 1994 Harley R. Myler and Arthur R. Weeks, "The Pocket Handbook of Image Processing Algorithms in C", PTR prentice Hall, 1993 Vishvjit Nalwa, "A Guided Tour of Computer Vision", Addison-Wesley Publishing, 1993 K. Siyan and J. Weaver, "Inside Java", New Riders Publishing, 1997
[The J Maker Home] [Program Home] [Pricing] [Contact] [FAQ]