A novel sub-part learning scheme is introduced in our work for the purpose of recognizing handwritten numeral images. The idea is borrowed from the concept of visual perception and part-wise integration of visual information by the cortical regions of the brain. In this context, each numeral image is divided into four half-parts: top-half, bottomhalf, left-half and right-half; the other half of the image being kept masked. An efficient data representation is derived in an unsupervised manner, from each image part, using convolutional auto-encoders (CAE), for our learning scheme that involves both early and late fusion of features.
If using this code, please cite our work using :
Susan, Seba, and Jatin Malhotra. "Learning image by-parts using early and late fusion of auto-encoder features." Multimedia Tools and Applications 80, no. 19 (2021): 29601-29615.
Link of the article: https://rdcu.be/cnHsd
Step 1: Preparing Patches (Top, Bottom, Left, Right).
Data Preparation.ipynb
Step 2: Learning feature vector for all patches.
Top.ipynb, Bottom.ipynb, Left.ipynb, Right.ipynb
Step 3: Obtaining early fusion probabilties by giving combined feature vector to classifier.
Early fusion probabilities.ipynb
Step 4: Obtaining probabilties score for each individual patch.
Late fusion feature.ipynb
Step 5: Fusing probabilties score obtained through early fusion and late fusion. Giving the fused vector to classifer to get the predicted class.
Late fusion feature.ipynb