Accessible image analysis for art assessment
The main purpose of this study is to illustrate capabilities of several public domain image analysis software (PDIAS) programs to complement subjective scoring, which is predominant in the field of art therapy. Although subjective scoring can provide a good deal of information the author feels quantitative approaches can provide additional data. Quantitative analysis is aptly suited for the determination of near-exact percentages and dimensions of formal elements, as shown in the results of the demonstration. However, the author maintains that a synthesis of quantitative and qualitative approaches is necessary for complete assessment, and thus emphasizes the addition of quantitative methodology. It is important to acknowledge that biases exist with quantitative approaches to collecting data. The discussion following the study explains some of these biases, particularly during the use of PDIAS. For this study, PDIAS programs were adapted for the analysis of the Face Stimulus Assessment to establish a model for how quantitative methods could generalize to other instruments. The software, explained in subsequent sections, includes Image Java (ImageJ), Scion Image, and Measuring Vegetative Health (MVH). National Institutes of Health (NIH) developed ImageJ to analyze biomedical images with high degrees of accuracy, which stemmed from comparing known to measured areas within an image. Scion Image is a descendant of ImageJ, and shares many of its capabilities. Educational programs and color analysis of satellite imagery are primary applications of MVH. The developer of MVHestablished is accuracy by testing select itemsprior to release, though independent testing is pending. Researchers using the latest computer technology developed these programs for distribution to colleagues at no cost.
The use of computer technology in art therapy is expanding and creating new avenues for research. The Discussion section explains potential developments that are workable by even the most novices of computer users, due in part to PDIAS. With computerized image analysis forming accurate rating scales, researchers developing common procedures can explore various techniques and terms from PDIAS. Digitally analyzing two dimensional art for assessment is accomplishable through PDIAS, as shown in the measurements portion of this study. As professionals begin to explore such programs, common procedures may develop from the research. Common procedures, theoretical approaches, and observational terms are all important characteristics of valid fields. Increased validity of computer-assisted art therapy assessment may result from standardizing processes of image analysis techniques. This and other studies have begun to explore objective methods of analysis applied to art assessment. Joining science and art could benfit art therapy by offering standardization and worth. This study incorporated practices and procedures from objective disciplines.
Standardized measurements overview
In this study, PDIAS measured several formal element categories featured in the FSA. These were color percentage, line, and shape (see Table 1). This analysis incorporated procedures from the development of the Computer-Color-Related Art Therapy Evaluation System. It also integrated research in image analysis and quantitative assessment.
Many art-based assessments exclude quantitative data. An overview of major art-based assessment tests revealed few with reasonable validity or reliability. To achieve credibility, art-based assessments should establish validity and reliability through standardization (Brooke). It is important to standardize measurement techniques so tests may be replicated; they should also measure what they claim. Developing standardized methods in assessment rating and scoring could assist in generating reasonable levels of reliability and validity through objectivity. There appear to be few objective solutions to problems arising from the interpretation of drawings.
Measurements
Color measurements
Recent technological advances allow for analysis of color. Human color perception varies according to each individual. Because of this variance, objective computer assistance in color rating warrants investigation. Color scoring for client insight is achieved through both manual and computer assistance in art-based instruments. In the Descriptive Analysis of Psychiatric Artwork, the presence or absence of red, yellow, orange, purple, green, blue, brown, white, and black was manually rated over 20 sections of each artwork. Three scales measured line thickness and color intensity. The raters from this study approximated form location.
Subjectivity poses problems for the analysis of artwork because the rater relies on judgment and observation. An early study attempted to distinguish phases of bipolar II based on several characteristics, including color; the inter-rater agreement from this study proved to be moderately accurate. A recent study revealed a low level of inter-rater reliability for the Prominence of Color in the FSA. Gantt and Tabone included a system for manually rating color in the FEATS. However, she also proposed a computerized scale for rating color categories.
Line measurements
Euclid defined a line as a length without breadth, and a straight line as something that lays evenly with all points on itself. Line is an infinite figure without dimension, extending in each of two directions. For art assessment, an assumption for this study is that line contains length and width. In addition, a drawing may contain numerous lines, leading to a line count technique, which is also a feature in many PDIAS programs.
Human perception of line is relative, especially when judging ratios and lengths. Human perception could lower inter-rater agreement. Comparing perception of line to objective computer analysis would therefore be a noteworthy approach to addressing relative perception. Furthermore, past research explored line and affective states, which is an additional factor to consider in the assessment of line.
Face Stimulus Assessment
Donna Betts constructed the FSA for nonverbal clientele exhibiting cognitive impairments. These individuals were unable to understand basic directives; therefore, the use of a stimulus drawing seemed a fitting means to elicit verbalization. Those with autism, communication disorders, and a-motivation responded well to face stimulus drawings. The design of this instrument was inspired by previous work in which research participant's significantly perceived vertical orientation of the face through their senses. The result was a series of stimulus drawings.
The FSA consists of three drawing sheets measuring the size of standard letter paper -- 21.59&times;27.94 centimeters (cm) each. The first is a pre-drawn face. The second is a featureless outline of a head. The third is a blank sheet of paper that assesses projective information.
Equipment
Analysis of the case study commenced with a flatbed Hewlett- Packard Printer-Scanner-Copier (HP PSC) 1410 with 4800&times;1200 dots per inch (dpi) resolution. Image processing involved a computer with 2.4 gigahertz (GHz) of read-only memory (ROM) and 512megabytes (MB) of random access memory (RAM).
Methodology
Processing
The basic steps to image processing include acquisition, discrimination, segmentation, measurement, and stereological interpretation. The final step of stereological interpretation normally pertains to three-dimensional objects, so this study excluded it. Acquisition, however, was necessary for image processing. Conversion to Tagged Image File Format (TIFF) from an analog state further readied the image for processing. The Joint Photographic Experts Group (JPEG) format, although high fidelity, does not allow for true image analysis.) Therefore, this study retained the TIFF format. Scale of the image format was set for accuracy at 84 pixels per cm, and filtering techniques lowered noise.
Filtering
Filtering is a component of processing that addresses noise. Noise can be introduced at various points of image acquisition and processing, and may alter the results of analysis. This analysis used a 3&times;3 median filter. The median filter is a common tool used in image analysis; it is also effective for image analysis applications. The median filter works by ranking neighboring pixels in a square, selecting the brightest pixel, and positioning it in the center of this square. The result is the rejection of certain types of noise. The C-CREATES analyses use this type of filter.
Case example
The individual in this case was a 46-year-old male who lived in a small rural area where he attended a voluntary community program sanctioned by a local mental health agency. He was unemployed, single, and formally diagnosed with schizophrenia by history. Several features within his drawing necessitated analysis. The author chose this particular drawing for analysis because it clearly revealed formal elements. The author also selected this case drawing because of bizarre features and lack of detail. Psychopathology of schizophrenic individuals is present in large collections of artwork. Individuals with schizophrenia generally included bizarre imagery and lack of detail in their depictions of the Person Picking an Apple from a Tree. In addition, yellow was prominent in the sun figure. In the DAPA, the yellow differentiated schizophrenia (p < .02) from other groups with an ANOVA level set at .05 (F = .02-.52, d.f. = 2, NS). There was also strict adherence to FSA manual guidelines during administration.
Color elements
The author broke down the drawing into color segments often rated in the FEATS, with computer methods contrasting with C-CREATES. Scale number 1 of the FEATS determined color use, whereas scale number 4 rated free space. The color white in this study was free space. While Kim et al. used expert systems, clustering, and edge length to separate color area, a method of thresholding tested by Murakami, Turner, Van Den Berg, and Schaberg was adapted for this study to produce the red and green percentages with Scion Image. Scion Image filtered the remaining colors prior to analysis with MVH. The number of highlighted pixels comprising each color generated percentages.
Results
Table 2 shows the results of the color analysis. The highlighted pixels representing color revealed blue as the color used least (.29%), while free space dominated the image (81.55%). Brown was prominent from numerous strands of hair drawn within the image. The color prominence second to brown was yellow, found in the sun ROI. The shape of this ROI approximated a circle with a roundness descriptor of .91 approaching the maximum value of 1.0. Counting tools analyzed 118 lines in the image. One-half of the total ends of the lines resulted in 59 drawn lines.
Discussion
With emerging technologies and the availability of PDIAS, the level of potential for research in both computerized procedures and art-based assessment is high. The main contribution of this research to the field of art therapy may be the development of highly accurate rating scales for art-based instruments. The results of this study mark an initial step towards this aim by giving the researcher a basic overview of PDIAS capabilities on the FSA.
Many issues concerning application, validity, calculation, and bias arose from using PDIAS to measure formal elements. Software engineers of three-dimensional imagery in the engineering and biomedical fields designed most of the functions. Therefore, additional research is required to make solid connections between these applications and art analysis. In addition, the FSA presents with further avenues of research. Analysis of the second drawing in the FSA series could reveal more diagnostic data about the client because there is less stimulus material to project upon.
Conclusion
The use of public domain image analysis software can be adapted to art-based instruments. The author explained common procedures in an image analysis so that interested professionals can embark on further research. The author used basic image analysis programs to generate objective data from a case study drawing. Case study results indicated that PDIAS is able to analyze accurately formal elements of an FSA drawing. This also illustrated the use of accessible programs. This procedure of image analysis applied to the FSA case study revealed potential for further research. The measurements included variations in the formal elements of color, line, and shape. Accuracy of these measurements stemmed from calibration. Objective information could yield figures more accurately than human raters could, though this claim warrants further research. Quantitative methods are one approach to unveiling assessment data. Even with computer technology, human raters are still required for inputting data, checking it, and assessing difficult material, such as bizarre content or emotional tone. Ultimately, both approaches are necessary for accurate assessment.
Existing computer analysis of art has not evaluated inter-rater reliability. Computerized image analysis has a relatively long history of application. It is widely used in science and art. Computer programs analyzed copperplate prints from the 15th to the 19th century. Historians entered original works into a computer through scanners, which aided in the preservation of artwork during analyses, though such studies did not include reliability measures. Increased sophistication in programming and the broadening of image analysis applied to art could result in better processing techniques. Such developments will permit scoring of less structured assessments, such as free drawings.