The School of Computer Science presents...
Augmented Color Input in CIR Requirements
MSc Thesis Proposal by:
Arnob Banik
Date: March 16th, 2026
Time: 1.00 pm
Location: 122 Essex Hall
Abstract:
Composed Image Retrieval (CIR) enables users to retrieve target images by combining a reference image with a text-based modification query. Although highly effective for structural or stylistic changes, it often struggles with precise color manipulation, because natural language is often too vague to capture specific chromatic preferences. This research proposes an extension to the CIR framework. It integrates a visual color input mechanism, a palette interface, alongside the standard image-text query. This provides an example of bridging the gap between subjective linguistic descriptions and objective visual intent, and an example of color personalization. The color information obtained from the user is mapped to the RGB color space and utilized during post-processing phase to filter and rerank the initial retrievals based on the dominant colors of the candidate images. The proposed module is designed to be plug-and-play, ensuring compatibility with both supervised and zero-shot CIR models without requiring extensive retraining of the base architecture. We will evaluate several implementation strategies to balance between accuracy and performance. This includes 3D color space indexing, a Gaussian Radial Basis Function (RBF) for similarity scoring, and a pre-filtering stage for low-latency response time in large-scale databases.
Keywords: Composed Image Retrieval, Multimodal Retrieval, Software Requirement Engineering
Thesis Committee:
Internal Reader: Dr. Ikjot Saini
Internal Reader: Dr. Muhammad Asaduzzaman
Advisor: Dr. Jessica Chen
Registration Link (MAC students only)
