Establishing haptic texture attribute space and predicting haptic attributes from image features using 1D-CNN.

Area of Science:

• Haptics
• Computer Vision
• Psychophysics

Background:

• Objective characterization of tactile textures is challenging.
• Existing methods lack a standardized framework for haptic texture representation.
• A standardized model analogous to RGB for colors is needed for haptic textures.

Purpose of the Study:

• To establish a standardized, four-dimensional haptic attribute space for texture surfaces.
• To develop a scalable model for representing and identifying haptic textures.
• To evaluate the performance of a 1D-CNN model in predicting haptic attributes.

Main Methods:

• Conducted a psychophysical experiment with human participants rating 100 real-life texture surfaces.
• Defined a four-dimensional haptic attribute space: rough-smooth, flat-bumpy, sticky-slippery, hard-soft.
• Trained a 1D-CNN model using psychophysical data and image features from texture samples.

Main Results:

• The 1D-CNN model demonstrated strong prediction accuracy for haptic texture attributes.
• The proposed model outperformed other machine learning and deep learning algorithms.
• The developed haptic attribute space provides a scalable framework for texture identification.

Conclusions:

• The established haptic attribute space offers a standardized method for texture classification.
• The 1D-CNN model provides a scalable and accurate approach for predicting haptic attributes.
• This research paves the way for objective and consistent analysis of tactile textures.