(Project) 3D Reconstruction of Textureless Surfaces

Muhammad Saif Ullah Khan, Muhammad Zeeshan Malik, Dardan Haliti

In this project, we want to reconstruct low-texture surfaces in 3D from a single RGB image. This task is non-trivial because low-texture surfaces lack distinctive features which makes reconstruction especially hard. We are working on developing a new neural network architecture for this task, collecting our own datasets, and training the network to generate accurate depth maps and surface normals of objects with low or no texture at all.

(Dataset) A novel segmentation dataset for signatures on bank checks

Muhammad Saif Ullah Khan
April 2021

The dataset presented provides high-resolution images of real, filled out bank checks containing various complex backgrounds, and handwritten text and signatures in the respective fields, along with both pixel-level and patch-level segmentation masks for the signatures on the checks. The images of bank checks were obtained from different sources, including other publicly available check datasets, publicly available images on the internet, as well as scans and images of real checks. Using the GIMP graphics software, pixel-level segmentation masks for signatures on these checks were manually generated as binary images. An automated script was then used to generate patch-level masks. The dataset was created to train and test networks for extracting signatures from bank checks and other similar documents with very complex backgrounds.

(Preprint) Investigate Cloth Artefacts of Loosely Coupled Sensor Networks with TailorNet

Muhammad Saif Ullah Khan, Bertram Taetz
January 2021

We investigate current techniques for modeling cloth deformation on human bodies in 3D, with a particular focus on the recent approach proposed by TailorNet. We introduce the representations for clothed digital humans in literature, and compare TailorNet to other recent data-driven models used for clothing prediction. We also talk about traditional physics- based simulation methods, and how they compare to data-driven methods in general and TailorNet in particular. Finally, we talk about fabric-embedded on-body sensors and how cloth artefacts can affect their readings. Use of body sensor networks to analyze human movement outside of laboratory environments is limited, because of the undesirable motion artefacts corrupting the movement signals. Inertial sensors worn on top of regular clothes are affected by the extra motion introduced by movement of clothes, making it is important to separate the body motion from cloth motion by identifying optimal locations for sensor placement, such that they are least influenced by the high-frequency wrinkles.

(Thesis) Signature Verification

Muhammad Saif Ullah Khan, Muhammad Mahad Tariq, Bilal Ahmad, Muhammad Imran Malik
May 2018

Signature verification is the process of using machine learning methods to validate the authenticity of an individual’s signature. Signatures can be of one of the two types; on-line or off-line, and this project focuses on off-line signature verification. Aim of this project is to design an algorithm which can distinguish between genuine and forged signatures using writer independent features, and to develop a system using this algorithm which can be used to verify signatures on bank cheques. We intend to build a complete end-to-end hardware/software system which can be used to acquire signatures from bank cheques, perform signature verification, and display the results. For this purpose, various deep learning techniques were developed and tested on standard datasets for off-line signature verification, as well as on a dataset collected by ourselves.