This paper presents a novel approach to interactive facial feature localization, aiming to achieve accurate segmentation of facial features on high-resolution images under various conditions. The authors propose an improved Active Shape Model (ASM) that allows greater independence among facial components and enhances the appearance fitting step by introducing a Viterbi optimization process along facial contours. They also introduce an interaction model that enables users to guide the algorithm toward precise results. The Helen Facial Feature Dataset, consisting of 2330 high-resolution, accurately labeled images, is introduced to evaluate the effectiveness of their method. The paper describes a component-based ASM (CompASM) that decomposes the global shape fitting into two modules: component shape fitting and configuration model fitting. This approach provides more flexibility for individual component shape variation and relative configuration between components, making it more suitable for handling images with large variations. The authors improve profile matching by introducing a new joint landmark optimization scheme using the Viterbi algorithm, which outperforms the standard greedy-based approach. They also propose an interactive refinement algorithm to minimize fitting errors for bad initial fitting. The authors evaluate their method on the Helen dataset and compare it with STASM. The results show that CompASM outperforms STASM by 16% on the Helen dataset. Additionally, the interactive refinement algorithm is shown to be effective in reducing errors quickly, with the most effective reduction achieved when both the linearly scaled movement and constrained refitting steps are used. The paper concludes that their approach is more robust for diverse face images and that further improvements can be made by learning from user behavior or guiding interaction through landmark uncertainty statistics.This paper presents a novel approach to interactive facial feature localization, aiming to achieve accurate segmentation of facial features on high-resolution images under various conditions. The authors propose an improved Active Shape Model (ASM) that allows greater independence among facial components and enhances the appearance fitting step by introducing a Viterbi optimization process along facial contours. They also introduce an interaction model that enables users to guide the algorithm toward precise results. The Helen Facial Feature Dataset, consisting of 2330 high-resolution, accurately labeled images, is introduced to evaluate the effectiveness of their method. The paper describes a component-based ASM (CompASM) that decomposes the global shape fitting into two modules: component shape fitting and configuration model fitting. This approach provides more flexibility for individual component shape variation and relative configuration between components, making it more suitable for handling images with large variations. The authors improve profile matching by introducing a new joint landmark optimization scheme using the Viterbi algorithm, which outperforms the standard greedy-based approach. They also propose an interactive refinement algorithm to minimize fitting errors for bad initial fitting. The authors evaluate their method on the Helen dataset and compare it with STASM. The results show that CompASM outperforms STASM by 16% on the Helen dataset. Additionally, the interactive refinement algorithm is shown to be effective in reducing errors quickly, with the most effective reduction achieved when both the linearly scaled movement and constrained refitting steps are used. The paper concludes that their approach is more robust for diverse face images and that further improvements can be made by learning from user behavior or guiding interaction through landmark uncertainty statistics.