دانلود ترجمه مقاله مدل ادراکی بصری دسته بندی برای طرح HCL

بخشی از ترجمه:

درک وبهره برداری از توانایی های انسان در سیستم بینایی بخش مهمی از طراحی رابطه های کربری وتصویری اطلاعات را قابل استفاده می کند. طراحی خوب برای درک سریع،آسان و از روی حقیقت از مولفه های کلیدی را قادر می سازد. جنبه های مهم چشم انسان توانایی آن برای درک بی دردسر وتبدیل آن به ویژگی های منحصر به فرد برای درک محیط وسازه ها واشیا است. ما برای درک مناطق گروه بندی وشباهت های ویژه منحنی هایی را برای مناطق منسجم آن در نظر گرفته ایم. در این مقاله ما در مورد یک مدل ساده برای طیف گسترده ای از ادراک پدیده گروه بندی،پرداخته ایم. یک تصویر دلخواه را به عنوان ورودی در نظر می گیریم وساختار توصیف شده آن در سیستم بصری را به صورت یک تصویر برمی گردانیم. این مدل نشان می دهد که می تواند از جنبه های قوانین طراحی سنتی مانند ضبط کردن وپیش بینی ادراک بصری در صفحه نمایش استفاده کند.
١-مقدمه:
طراحی یک رابط کاربریو گرافیکی برای درک ،اطلاعات کمی دارد وتا حدودی از نظر اثربخشی ضعیف است. تعدادی از مسائل را تحت تاثیر طراحی هستند ،زمینه های رفتار انسانی می تواند به وجود بیاورد. طراحی باید یک شناخت خوب باشد. یعنی کاربر ساختار معنایی طراحی را به راحتی درک کند.تفسیر اطلاعات بصری می تواند در طراحی با زحمت زیادی انجام شود. در اینجا ما برروی جنبه های ادراکی طراحی تمرکز می کنیم. شاید مهم ترین جنبه چشم انسان برای طراحی،سیستم ادراکی آن است. سیستم ادراکی ،اشاره به پدیده هایی دارد که در آن سیستم بینایی به سرعت و به ظاهر بی دردسر و تخمینی از ویژگی های منحصر به فرد را از درک منسجم محیط ،سازه واشیا دارد. این پدیده برای اولین بار توسط روانشناسان گشتالت مورد بررسی قرار گرفت که مجموعه ای از موردهای مطالعه به صورت کیفی براساس اصول حاکم بوده است. از جمله درک این الگوها که به صورت محدود نیت به شرح زیر است:
تمایل به چیزهایی که در نزدیکی به گروه هستند(قانون نزدیکی گشتالت)
به اشتراک گذاشتن ویژگی های مشابه(قانون شباهت)
مستمر بودن (قانون ادامه)

بخشی از مقاله انگلیسی:

ABSTRACT Understanding and exploiting the abilities of the human visual system is an important part of the design of usable user interfaces and information visualizations. Good design enables quick, easy and veridical perception of key components of that design. An important facet of human vision is its ability to seemingly effortlessly perform “perceptual organization”; it transforms individual feature estimates into perception of coherent regions, structures, and objects. We perceive regions grouped by proximity and feature similarity, grouping of curves by good continuation, and grouping of regions of coherent texture. In this paper, we discuss a simple model for a broad range of perceptual grouping phenomena. It takes as input an arbitrary image, and returns a structure describing the predicted visual organization of the image. We demonstrate that this model can capture aspects of traditional design rules, and predicts visual percepts in classic perceptual grouping displays. Author Keywords Perceptual organization, grouping, good continuation, proximity, similarity, Gestalt, contour integration. ACM Classification Keywords H5.2. User interfaces, theory & methods. INTRODUCTION Design of user interfaces and information graphics is poorly understood, and somewhat hit-or-miss in terms of effectiveness. A number of issues influence the success of a design, and these run the gamut of the underlying human behavior. A design must be good cognitively (can the user easily understand the semantic structure of the design?), perceptually (can they effortlessly interpret the visual information present in the design?), and socially (does the design fit into the user’s workflow? will they want to use it?). Here we focus on perceptual aspects of design. Perhaps the most important aspect of human vision for design is perceptual organization. Perceptual organization refers to phenomena in which the visual system quickly and seemingly effortlessly transforms individual feature estimates into perception of coherent regions, structures, and objects. These phenomena were first studied in detail by the Gestalt psychologists, who produced a set of qualitative Gestalt principles that govern pattern perception [1, 2], including but not limited to: the tendency of things to group if they are nearby (the Gestalt law of proximity); if they share similar features (the law of similarity), or are smooth and continuous (the law of good continuation). The duals of perceptual grouping are important phenomena in their own right: we quickly and effortlessly perceive boundaries between certain visual textures, perceive edges between coherent regions in an image, and quickly detect unusual items that seem to “pop out” from the background. Examples of perceptual grouping phenomena are given in Figure 1. Following the visual system’s “rules” of visual organization makes interpretation of visual aspects of designs effortless: a user easily sees which labels refer to which parts of a diaa b c d e f Figure 1: Perceptual grouping examples, including grouping by proximity & similarity (a, b), and grouping by good continuation (c, d). (e) A user interface; what is the percept? (f) A graph, from [5]. Will a user perceive the trend of the data? Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CHI 2009, April 4–9, 2009, Boston, MA, USA. Copyright 2009 ACM 978-1-60558-246-7/09/04…$5.00. CHI 2009 ~ Cognitive Modeling and Assessment April 8th, 2009 ~ Boston, MA, USA 1331 gram, notices a trend in data, and makes connections between the “overview” and the “detail” in a map. Good designs use the natural perceptual processing power of the brain, and interpretations of such designs are fast, robust to instruction, and cross-cultural [3]. With poor visual design, the grouping structure may not match the structure of the information, leading to confusing displays [for examples, see 4, 5]. A user might see columns of information where in fact there were intended to be rows; incorrectly group two regions of a graphic that have no relation to each other, and so on. In Figure 1f, do the +’s at the grid points interfere with perceiving the data curves? However, models for how users extract meaning from visual displays are incomplete. Designers often “eyeball” it– i.e. try to judge perceptual groupings using their own visual systems, perhaps using tricks such as “squinting” at the design or viewing it from far away to bring out coarser scale groupings. Various researchers have suggested general guidelines for design [3, 5, 6, 7]. Many existing models in use in the HCI and information graphics field are specific to particular types of displays e.g. alphanumeric [8], text documents [9, for a review], visual basic dialog boxes [10], and sketch editing with vectorized input [11]. In addition, some work has been done to translate very basic rules of thumb about “preattentive” (i.e. fast and effortless) visual processing to the design of visualizations [12]. Rules of thumb, based on simple behavioral experiments, are useful in understanding and guiding design. However, designers may have difficulty applying them to more complex displays. Ideally, one would prefer a model that could predict the likely perceptual groups for an arbitrary design. Such a model would be most useful if its mechanisms and output were easy to understand, as this transparency would aid a designer in making changes to a poor design. In this paper, we draw on tools from statistics as well as recent work in computer vision to propose a model of perceptual grouping that is simple to understand and implement, yet effective. This model can predict image segmentation, contour integration, segmentation of orientationbased textures, grouping by similarity and proximity in standard Gestalt displays, segmentation of natural images, and grouping in more complex diagrams. PREVIOUS WORK A great body of computer vision work exists on the topic of perceptual grouping. However, much of it is inadequate for user interface (UI) designs and information graphics. After many years of human and computer vision research, results of image segmentation models are still often quite poor. For one thing, many of these models have inherent biases against extended groups. A classic result is the tendency to predict the perceptually invalid segmentation of a pure blue sky into 3-4 separate regions. Furthermore, the better models often do not easily lend themselves to intuitions. For example, the normalized cut algorithm [13], which works fairly well, treats image segmentation as a graph partitioning problem, with each pixel a node. The similarity between two pixels determines the weight between the corresponding nodes. This weight can be thought of as the tightness of a spring. The algorithm partitions this spring-mass system into regions that will tend to move independently. This is an evocative description, but it does not lend itself to easy intuitions about the predicted segmentation, nor how one might change a display to obtain a different grouping. In information graphics, often interesting groupings form between non-physically adjacent items. We want to know, for instance, whether a user will easily perceive the association between colored lines on a plot and the colors in a legend. Will it be obvious to a user that a set of buttons on a remote control perform related functions (Figure 1e)? The vast majority of computer vision algorithms group only contiguous regions. It is unclear how well these algorithms can extend to group over gaps between items. This is a serious problem for applying these algorithms to UI designs. Contour integration algorithms do group across gaps [14, 15, 16, 17], but have not been extended to grouping based upon other Gestalt principles. Techniques that cluster in luminance or color space, for example k-means and non-parametric equivalents [18, 19], will group across gaps in space. However, these techniques, as commonly used, take this loosening of the proximity constraint too far; they will tend to group independently of proximity.Another difficulty is that typical perceptual organization models do not produce a hierarchical grouping, though see [20, 21, 22, 23, 9]. Something like a hierarchical percept clearly exists [24, 25]. In a text document, for instance, individual letters group to form lines of text, which form paragraphs, which form columns, and so on. There has been little work comparing models of perceptual grouping with human perception, though for partial attempts see, for example, [26, 27, 28]. In part this is due to the lack of quantitative data on perceptual organization. More seriously, though, many of the algorithms perform too poorly to predict even what qualitative data exist. Predicting a wide range of simple qualitative perceptual organization behavior would be a significant step forward. A more subtle difficulty with existing models of perceptual grouping is that nearly all handle only one kind of perceptual grouping, e.g. they only find boundaries between regions of natural images [20, 21, 22, 26, 29, 30, 31, 32, 33], perform grouping by proximity [23, 9], segment textures [27, 34, 35, 36, 28], or do contour integration [14, 15, 16, 17]. Certainly if separate models for the different grouping phenomena were required, we would use separate models. However, we will demonstrate that this is not necessary. Advantages of a single, unifying model include the fact that the outputs are compatible, and thus it will be easier to combine them to get an overall picture of the perceptual structure of a display. Furthermore, the difficulty of getting CHI 2009 ~ Cognitive Modeling and Assessment April 8th, 2009 ~ Boston, MA, USA 1332 intuitions about what will group is significantly reduced if there is essentially one model to understand instead of four. The brain itself contains the best perceptual organization system in existence, and it is generally believed to use similar mechanisms to perform related tasks. More recent models have become more unified largely due to a unified vision of the purpose of perceptual organization. (Although see, for example [37], which captures a number of perceptual organization phenomena based on a unified view of the underlying neural operations.) This vision echoes that of Helmholtz [38], who argued that what we perceive is our mind’s best guess of what is in the world, based on summary data (the input image) and prior experience. By this argument, the goal of perception is to create what the computer vision field refers to as a “generative model” for an image: what processes created the image, and where each of the processes operates. Perceptual grouping concerns itself with the latter. Based on this view of perceptual grouping, a number of computational models have been developed for image segmentation [39, 13], edge detection including texture segmentation [28, 26, 36, 40], contour integration [41, 42, 43, 44] and distinguishing figure from ground [45]. Our model similarly finds groups by attempting to infer the processes that generated the display. THE MODEL We motivate our model in detail, since the intuitions to be gained apply to design as well. The Representation is Key Why is it difficult to predict perceptual groupings? Many perceptual grouping problems appear difficult to explain via simple processing in the image domain. Consider the Gestalt disk array example in Figure 1a. The typical percept in this figure is of an array of disks with three groups: the gray disks on the left, the white disks on the right, and the background. It is this grouping we hope to mimic. In determining what regions form groups, both proximity, i.e. difference in (x, y), and similarity, i.e. difference in feature (here, luminance) value, are relevant. Virtually any simple algorithm can group the pixels in each individual disk, since they are touching and contain pixels with exactly the same luminance. The difficulty comes from variations in luminance, and bridging the gaps between individual disks to group them in spite of the lack of spatial adjacency. However, spatial information is also important – you do not want to group together all pixels of the same luminance, regardless of how far apart they are.