1.1 Related Research and Theory


The ability to search or query a database of images is a necessary andrequired functionality inthe design of the interface of image databases. The query feature isneeded in order toaccommodate users who wish to find a particular image from among themany images present inthe collection.

One common method used for querying image databases is to allow the userto enter a textdescription of a target image. A computer algorithm then uses this textdescription to match thequery with text descriptions of images in the database and outputs thepossible matches. However, this method contains intrinsic problems due to the complexitiesinvolved in entering atext description of a highly conceptual image. In addition, thealgorithms which perform thequery are hindered by ambiguities in human written language; thereforethey are likely toproduce multitudes of incorrect matches. More advanced methods ofsearching an imagedatabase have arisen in order to rid the need for text descriptions ofimages. One veryconceptual method is Spatial-Query-By-Sketch (Egenhofer 1996).Spatial-Query-By-Sketchrequires the user to draw a simple sketch of the image sought. It thenexecutes a complexalgorithm to generate matches for the query based on the input of theuser sketch.But, while this method does provide the user with a more direct modelfor searching, the userwill still eventually be left with a series of possible matches, someof which are incorrecttranslations of his/her sketch.

As a result, the need for a well designed and efficient image browser isapparent. If a query isnot performed, the image browser is the only method of searching a givendatabase. Additionally, if a user does perform a query, the user must alwaysbrowse through the possiblematches given by his/her search criterion, since even the most advancedmethods of queryingavailable today are not exact. Also, query algorithms can sometimeseliminate (and not display)images the user might be searching for; a browser is needed in this casein order for the user tofind such images.

This study which measured the efficiency of several types of browserswas driven by this necessity, and consequent popularity of image browsers in today'ssoftware market.

Image Browsers And Computer-Human Interaction

In "Image-Browser Taxonomy and Guidelines for Designers" Plaisant, Carrand Shneidermanstated that in contemporary image browsers, the"one-dimensional scrollbar is a well-establishedfixture" (1995). The paper also suggested that the core functionalityof the one-dimensionalscroll bar allows the user to not only scroll through a document, butalso to jump through thedocument in increments. Plaisant et. al. contended that mostdesigners prefer to take advantageof the widespread familiarity with the scrollbar interface and so usethis particular interface intheir designs.

The two one-dimensional scroll bar functions of scrolling and jumping inincrements wereexamined separately in this experiment. One of the browsers was capableof jumping inincrements while a second browser used scrolling to maneuver through thedatabase.

A third type of browser, titled the "Velocity Slider" was also used inthis experiment. TheVelocity Slider is unique in that it allowed the user to control thespeed at which the displayedcolumns of image thumbnails shift off the screen. A bar was present atthe bottom of the screen,labeled incrementally from -5 seconds to +5 seconds. A selection ofnegative time shifted thecolumns of thumbnails to the right (in increments from 5 seconds to 0.5seconds, depending onthe user's selection) while a selection of the positive time shifted thecolumns to the left. Theuser was free to adjust the timing of the shifts (and hence the speed atwhich the columns ofthumbnails moved by) at anytime during the process. The Velocity Slidereffectively took awayhuman control (except for speed regulation) from the increments of thepages of images.


In an investigation of thumbnail image sizes versus user response time,Liebeskind, North andOrandi examined the time to find a target thumbnail image amongdistractor images, in gridsets of 1X1, 3X3 and 6X6 thumbnail layouts (1994). This experimentshowed that as the size ofthe thumbnails decreased, (and so the total number of thumbnailsdisplayable increased) thetime required to find the target decreased.

In "Visual Search in Continuous, Naturalistic Stimuli," Wolfe found thatthere is a linearrelationship between the size of a target object (surrounded bydistractor objects), and the timefor the user to find the target object (1994). More specifically, thetime to find the targetincreases as the size of the set of items increases (as the absolutesize of the target objectbecomes smaller). Wolfe found this to be true using naturalistic stimuliie. objects anddistractors which are natural objects viewable in everydaycircumstances.

Wolfe's result along with that of Liebeskind et. al. suggest that thereis an optimum thumbnailsize such that the quality of image recognition is balanced with thetotal number of imagesshown at a single time on the screen.

Due to these effects, two different grids (3X3 and 5X5) of thumbnailswere used in conducting this particular experiment. Users using the 3X3 grid hence viewedlarger thumbnails than thosewho viewed the 5X5 grid. These two grid sizes were used since theyroughly estimate theboundaries of the optimal thumbnail size. This was done in order toestablish the independenceof the thumbnail size effect on this experiment which determines thequality of the three typesof image browsers. In short, the same relative results were expectedfor each browser regardlessof thumbnail size.

Furthermore in "Local and Global Factors of Similarity in Visual Search"von Gunau, Dube andGeleria proved that the more similar the target and distractors are toeach other, the more time itwill take to find the target (1994). In this experiment, the targetimages were chosen to be asdistinct as possible when compared to the entire database; this was doneto minimize the effectfound by von Gunau et. al.

Finally, some thought was given to the psychological facts given in"Parallel Verus SerialProcessing in Visual Search" (Egeth and Dagenbach 1991). This paperestablished that inbrowsing a set of images, humans can use either parallel or serialprocessing. Parallelprocessing involves a scenario where the user searches the images onscreen by taking anoverview of the screen and the target image becomes instantlyrecognized. Serial processinginvolves a process where the user looks at the pictures on the screenone by one seeking thetarget. However, Egeth and Degenbach also stated that it would be verydifficult to tell whichmethod a subject used and if he/she changes method during theexperiment. Therefore, it wouldbe futile to attempt to isolate parallel vs. serial searches in thisexperiment. Howeverconsideration must be given to the fact that drastically differentsearch times can arise from theuse of serial vs. parallel searching.