Go to previous article
Go to next article
Return to 2005 Table of Contents
Stephen E. Krufka
Email: krufka@udel.edu
Tactile imaging can aid the visually impaired and blind in accessing information from various types of media, for example graphics. Unlike a verbal description of an image, tactile pictures better convey spatial information, for instance in maps where tactile representations improve geographical education and mobility for the blind. Graphical information from charts, graphs, flow charts, and diagrams are also better represented by tactile graphics. With the increasing dependency on graphic user interfaces, tactile imagery can provide accessibility to computers and the World Wide Web.
Until recently, the production of tactile pictures was a manually intensive process. Raised-line drawing boards could be used to create raised lines on a plastic film. Tactile-experience pictures where composed of a variety of materials including wood, sandpaper, and cloth glued to a cardboard base. More modern methods have evolved including the production of Braille graphics. A standard Braille printer can automatically map computer images into Braille dots, making it a relatively fast and effortless process.
Many Braille printers have the ability to emboss Braille dots up to a 20x20 dpi resolution with varying Braille dot heights to represent gray scale, texture, or even color. One solution used to create tactile pictures is to emboss the image directly. The result is a lower resolution tactile conversion where dark and light image regions are represented by high and low Braille dots respectively. This method is effective when dealing with simple pictures. For more complex images however, this method yields too much information to be comprehendible by touch resulting in output that is confusing.
An automatic visual image to tactile conversion method is presented in this paper. The method produces simplified tactile edge-maps that contain only object boundaries, thus representing the fundamental information in an image without introducing non-essential details. Within this edge map significant edges are embossed with higher Braille dots to bring more attention to them. In particular, the method focuses on vector graphic formats that have shown great potential in providing accessibility to the blind [1].
Unlike raster images where an image is defined by a set of pixel intensity/color values, a vector graphic contains a set of fundamental shapes with various parameters, e.g. circles, squares, and arbitrary shapes with various sizes, placement, fill color, outline color, and other attributes. By combining these fundamental shapes, a vector graphic with any level of complexity is created. When a user views the graphic, the shapes are processed to form a viewable bitmap.
The shapes within a vector graphic can be arranged into a hierarchical tree. This tree is instrumental in determining what boundaries are significant. The hierarchical tree has the following properties.
As a result of the last property, the trees ancestor nodes maintain the shapes with the outer-most outlines (outlines enclosing other outlines), while the descendant nodes maintains the outlines inside others. Outermost outlines generally represent major objects, while the inner outlines are parts or details of those objects. The depth of a node is equal to the number of shapes that it is contained within.
This structure offers considerable advantages over raster formats in the production of tactile edge maps. First, all potential object boundaries are contained by the collection of all shape outlines. These outlines can be extracted; however extra measures are still needed to ensure an outline is a true boundary (discussed in the Edge Extraction Section). Second, the hierarchical tree structure of vector graphics enables one to find the significant edges in a graphic as discussed in the Edge Significance Section.
Research [2] has indicated that humans use the same principles when exploring an outline picture by touch or by sight, leading to the conclusion that any system that represents boundaries of surfaces is useful. Way [3] showed that by applying boundary detection (by means of Edge Detection or Segmentation) in the generation of tactile pictures, ones ability to discriminate, identify, and comprehend them improved greatly. The boundaries of objects within an image are often manifested by sharp color transitions commonly referred to as edges. By extracting the edges, the boundaries are obtained.
With vector graphics, an edge map can be obtained by simply replacing all shape fill colors with the background color (generally white) while making the outline colors another color. To keep depth data (corresponding to hierarchical tree), the outline color for each shape is set to its own depth. After converting the altered graphic to a bitmap, an edge map is obtained where an edges intensity is defined by its depth.
This extracts a large set of possible boundaries, which unfortunately may include false boundaries. For instance, color-fading effects are produced through overlapping shapes (of similar color), causing false edges. Outlines within these regions should not be included since color transitions are generally slight (low contrast) and do not represent edges.
In a comparison between visual and tactile graphics, Kurze [4] observes that important lines in tactile images should be enhanced to bring more attention to them. Redundant objects and details should be removed to provide clarity of the images defining features. This provides motivation to determine the significance of an edge so the subjects' attention is focused on the defining edges.
In general the outer boundaries (boundaries that surround other boundaries) are of great importance. Using this logic, distinct objects (not within other objects) will be emphasized, thus stressing the importance on image objects as a whole, but not the details or parts of those objects.
To remove or de-emphasize details, the hierarchal tree is pruned at some given depth. Shapes having depths greater then some selected depth are deemed ambiguous, while lower depth shapes are deemed significant. By performing this pruning operation, the edge map is altered such that higher and lower depth edges are labeled as ambiguous and significant, respectively.
Although pruning the tree is an effective method for assigning significance, having a user-defined depth setting is not desirable since vector graphics have varying levels of complexity. One graphic may have a high maximum depth while another is considerably lower, leading to a setting that works well for simple graphics, but not for complex ones.
To find an appropriate depth setting, an algorithm that adapts to various graphic complexities is presented. A histogram from the edge map is obtained to provide a measure of the total number of edge pixels for each depth level. The user can enter what percentage of the edge map they wish to deem significant. By using this percentage setting and the histogram, a depth setting is found such that the percentage of edge pixels deemed significant is closest to the user setting. The depth setting found by the algorithm is then used with the pruning operation.
The user defined percentage setting allows the user to tailor their tactile pictures to their own preferences. By selecting 100%, the user decides to stress the same emphasis on all lines. Selecting 0% minimizes the emphasis to only the outermost outlines, which provides a tactile graphic with an outlining reference for each distinct object. Finally, an intermediate selection can be made to enhance more edges within the tactile picture. The Results section illustrates the different selections.
Most visual images have a much finer resolution then the tactile sense can perceive. The tactile sense is limited to roughly 20x20 dpi resolution [3], where the edge maps processed in our case are 100x100 dpi. Hence the resolution of an edge map must be lowered to the tactile resolution prior to embossing, a process called downsampling.
After downsampling, the edge map is sent to the Braille embosser. Significant edge pixels are represented with high Braille dots while other edge pixels are created with lower Braille dots.
The figure illustrates an example of a vector graphic processed by the proposed algorithm with various edge percentage settings. Black and gray squares are used in the examples to represent significant and ambiguous Braille dots respectively. Significant Braille dots are embossed with a greater height then ambiguous ones in the final tactile picture. (a) shows the vector graphic while (b), (c), and (d) shows the output with the percentage setting equal to 0, 0.5, and 1, respectively.
Go to previous article
Go to next article
Return to 2005 Table of Contents