PalmHashNet: Palmprint Hashing Network for Indexing Large Databases to Boost Identification

Abstract

Palmprint identification aims to establish identity of a given query sample by comparing it with all the templates in the database and locating the most-similar one. It becomes compute-intensive as the number of comparisons becomes proportional to the size of the database. The process needs to be fastened to get response in real-time especially for large databases. This paper proposes a palmprint database indexing approach called PalmHashNet that generates highly discriminative embeddings to create a fixed-size candidate list for comparison to make identification a constant time operation. Acquired palmprint images are fed to the feature extraction network which is pre-trained using softmax loss. To minimize the intra-class distance between samples belonging to the same class a margin is added to the softmax loss. This ensures that the features have high intra-class and low inter-class similarity. k-means and locality sensitive hashing (LSH) are explored for index table creation. In this setting, cluster centers for k-means and hash values in case of LSH serve as indices. For a given query palmprint, the features are extracted and compared with the index values. The candidates lying in the most similar bin are retrieved for identification. The proposed approach gives probabilistic guarantees for the query to be identified in the selected bin. Experiments are conducted on four widely used palmprint databases viz. CASIA, IITD-Touchless, Tongji-Contactless and Hong Kong Polytechnic University Palmprint II (PolyU II) database. The proposed approach achieved a penetration rate of 0.022%, 1.032%, 4.555% and 0.39% at 100% hit rate thus, making the identification process approximately 4500, 96, 21 and 256 times faster on the considered databases respectively. The advantage of the proposed approach is that the query palmprint needs to be compared with only a small percentage of the database instead of the complete database.

Efficient palmprint biometric identification systems using deep learning and feature selection methods

“…The approach is tested on three palmprint databases IITD-Touchless, GPDS-Touchless, and CASIA. Arora et al [32] introduced PalmHashNet, a novel indexing method that learns compact feature vectors for palmprint identification. They used the Softmax loss function with additive margin to train the model to index the palmprint database and to simultaneously learn the feature vector embeddings.…”

The Complexity of the Data Retrieval Process Using the Proposed Index Extension

“…Sequential data block loading, scanning, and evaluation are removed, replaced by the proper data block identifying where the relevant data reside. Whereas the index size is significantly smaller than the whole table [14], memory block loading necessity requires fewer blocks. Thus, the correct indexing strategy is crucial for achieving maximum database performance.…”

“…Based on [18,33], additional demands can be applied only in specific situations when the original data row could be located in a partially free block, which is currently loaded into the memory during the evaluation. It generally reflects less than 1% additional demands, reflected by the I/O block loading necessity [14]. The remaining free space in the blocks is used for the update operations.…”

“…Generally, it can be defined by various structures, consisting of the ROWID value, used as a direct locator pointing to the database, file, block, and position inside the block. Using the index, sequential data scanning is reduced to the index search and block identification using ROWID [14,15]. The finest granularity of the processing is the block itself, which must be fully loaded into the memory for the evaluation and consecutive processing.…”

“…The suitability of the index is crucial for the processing, whereas the order of the attributes delimits the index structure. Based on the [14], the order of the elements inside the index should reflect the query conditions and optionally values listed in the Select clause. If all required values (attributes and function calls related to the table) are not present in the index, using the ROWID values, the whole row can be easily located and memory-loaded.…”