Contributions and Limitations

Even though many researchers agree that color is highly useful in image segmentation during the early visual processing (Callaghan, 1984; Gegenfurtner &

Rieger, 2000; Li & Lennie, 1997), the role of color in high-level vision is still a controversy in the literature. Some claimed the facilitation of the color in object representations while some others argued no role of the color in object representations (for a review, see Tanaka, Weiskopf, & Williams, 2001). To investigate the effect of

118

the color in objects representation, previous researches manipulated the color of the objects as the appropriate, inappropriate and monochromatic colors to measure their effect on recognition times. However, three reasons prohibit us to draw a definite conclusion from those studies. First, the chromaticity and luminance components of their stimuli were not well separated. The so called "color stimuli" is usually composed of both chromatic and luminance components. The chromaticity and luminance are mediated by different post-receptoral mechanisms and their effects do not necessary sum linearly. A stimuli contained both components may produce quite different effects in different channels. Hence, a comparison between the color and monochromatic effect makes little sense. Second, the number of the color-selective channels that involve in the so called color conditions was ill defined. The number of the channels involved may influence the performance. Hence, it is a possibility that the inconsistency among previous research is due to the stimuli they selected. For the stimuli that contain more colors, the color showed facilitation effect on object representations while for the stimuli that contains less color, the color did not show facilitation effect. Third, the stimuli in these studies were the objects familiar to the observers, such as a banana or vegetables. The observers have a prior knowledge of their form and color. Color might provide useful information for the recognition of some, but not all, objects (Tanaka, et al., 2001). For example, although ‘red’ might be an informative cue for identifying fire engines, but is certainly not a very useful cue for identifying automobiles or bicycles. Similarly, the form of some objects might be easier to recognize so that the effect of the color is hard to observe. Hence, the effect of the color depends on the selected objects. The results make sense only when the researchers control the observers’ form and color knowledge to each object. Recently, the research has noticed the color knowledge issue while still ignore the form

119

knowledge issue.

In this thesis, we use symmetry as a tool to investigate the role of color in high-order form perception. The advantage of using symmetry as the stimuli is preventing the influence of the prior knowledge of both form and color. In addition, we selected the colors on the three cardinal axes of the color space to separate the chromaticity and luminance. This also controls the number of the color-opponent channels involved. Our results should more appropriately demonstrate the role of color in high-order form perception. As summarized above, our results showed that given the same number of the colors in the images, the chromatic colors do not facilitate higher-order form detection comparing to the achromatic colors. However, the more colors in the images, the easier to detect symmetry of them regardless of chromatic or achromatic colors. Our model can fully account for the above results. This helps us further understanding the possible mechanisms behind.

Even though our manipulation avoids several pitfalls in the previous studies concerning the higher-order color vision, some might still argue that our results cannot generalize to the real-life objects or the nature scenes for our stimuli were rather abstract and looked unlike a real life object. After all, our stimuli were composed of sparse random dots while the objects or images in the nature scenes contain denser texture. In addition, the colors in our stimuli distributed randomly while those in the real-life objects or scenes tend to be well organized. We however consider that our results should have no problem to generalize to the real-life objects or scenes. First, the increment of the number of the colors facilitates symmetry detection by reducing the inhibition term of symmetry channels. That is, the more colors in an image, the less inhibition term the symmetry channels have. The inhibition term is proportional to the density of the image components (either

120

symmetric pattern or noise). Hence, as long as the increment of the overall density does not change the density relationship among the images contain different number of the colors and in turn their inhibition strength to the symmetry channels, our results would have no problem applying to the dense texture. Second, while the colors in our stimuli distributed randomly and thus were not well-organized, we can still observe an increase of facilitation effect to the symmetry channels as the number of the colors increases. Hence, we can expect that this effect is even more obvious in the real-life well-organized objects or scenes.

However, our results did show the influence of the prior knowledge on high-order form detection. For example, the mechanisms of the facilitation of the number of the colors on symmetry detection differ when the observer has and has no prior knowledge of the axis orientation of the symmetric patterns (Chapter 7 and 8). Hence, it is likely that the prior knowledge of the color in the images or objects influence the detection performance. In this study, we did not manipulate the prior knowledge of the color. The mechanism of the prior knowledge of the color in symmetry detection is unclear. However, based on the current understanding of color-form integration in this study, we can further investigate the influence of learning, experience, and knowledge on chromatic form detection in the future, to get a full picture of higher-order color and form vision.