Color-based object recognition: Analysis and application

• Main Author: Reyes, Napoleon H.
• Format: text
• Language: English
• Published: Animo Repository 2004
• Subjects: Colors; Computer algorithms; Fuzzy logic; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_doctoral/940; https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1939&context=etd_doctoral
• Institution: De La Salle University

A novel indoor color-based object recognition algorithm is presented in response to the challenges posed by the current machine vision research status. Previous researches found that current color constancy technique are inadequate in dealing with color object recognition. Furthermore, researches in the field of fuzzy multi-channel color imaging are rather sparse. Contrary to color constancy algorithms, the algorithm develop focuses on manipulating a color locus depicting the color locus depicting the colors of an object (mobile robot), and not stabilizing the whole image appearance per se. The rationale behind the development of the new color object recognition algorithm is inspired by recent findings in the neurophysiological aspect of the human visual system, which suggests that contrast computation precedes segmentation. This research contributes in the field of Color Science by providing a new set of color descriptors that adheres to human perception of color, in a new transformed rg-chromaticity space. Moreover, this research extends the computing prowess of fuzzy logic in the realm of multi-channel color imaging by providing a new breed of gradually adaptive multi-channel fuzzy color inferential filter for color locus constancy, called Reyes-Dadios Color Contrast Fusion (RDCCF). RDCCF utilizes a new color contrast degrade operator that was originally mathematically derived in this work, along with color contrast enhance. RDCCF is the first of its kind in the family of true fuzzy inferential filters for tracking down objects via color. The RDCCF algorithm locks on a color locus depicting the target object, whilw accurately compensating for the effects of glare, and hue and saturation drifts. Experiments on color spotting similar (hue-related) colors show that the new color decriptors and RDCCF combined, is better than using rectangular and pie-slice decision regions in UV space. In addition, this research provides an algorithm for recognizing mobile robots even in the event of collisions. Also, this work presents an algorithm for predicting the position of the ball. Empirical results of the application of the color-based object recognition algorithm on the robot soccer game attest to its robustness under spatially varying illumination intensities, multiple light sources, presence of highlights, object rotation and collision, in real-time.