A Dominant Color Extraction from Images/Videos for Mobile Devices

Image and video dominant color extraction is the most important use of image processing, but it misses some of the more important colors in smaller or less common areas. In this paper, a new method of dominant color extraction and classification is proposed; it circumvents the issues in previous methods by leveraging the effectiveness of $\mathrm{U}^{\mathrm{2}}$ Net Lite-based foreground-background separation and the ease of K-means for color patch detection. Our approach starts with segmenting the input video or image with the help of the $\mathrm{U}^{\mathrm{2}}$ Net Lite model into distinct regions to enable region-specific analysis for foreground and background, following which we independently perform K-means clustering on both foreground and background to detect color patches in order to pick up major colors in each segment. These patches of colors are then combined and classified into distinct classes of colors, and this gives a comprehensive and true color representation of the whole image. Results Experiments indicate that our method not only recovers the most frequent colors but also captures distinctive colors in a limited area, therefore enhancing the accuracy and reliability of color analysis in compound images. Thus, this approach has a lot of application value in visual content analysis, object recognition, and multimedia retrieval.