Assignment Summary of Article Adversarial Cross-Modal Retrieval

   In this article, the authors have focused on the adversarial of cross-modal retrieval to have flexible experience of retrieval across a variety of modalities such as images v. texts. It has been explained that research related to cross-modal retrieval has been associated with the learning of subspace, in which a comparison is made between different modalities. To understand the perspective, a method of novel Adversarial Cross-Modal Retrieval (ACMR) has been presented. The role of adversarial learning is critical in this regard, and this is what has been the focus of the researchers that they used adversarial learning between the given process, and its role was interplay in this regard. It is vital to understand that when the purpose is to benefit from the multimedia data’s abundance, then multimedia technology can be used for more optimal performance. The automated mechanism can play a vital role in this regard. In this research, it was important to understand the complete mechanism of Adversarial Cross-Modal Retrieval so that maximum output is taken from the research to improve the process in so many ways. The first vital thing in this research was to develop a method. After analysing the article, it was evident that proposed. In the proposed method, different elements were analysed one by one such as Adversarial Cross-Modal Retrieval, Modality Classifier, Feature Projector, as well as, Problem Formulation.

            Different methods worked under different elements. The feature projector was the first process, where a modality-invariant representation was generated in a given common subspace. This process is confused with various other processes such as modality classifiers, which is used to make discrimination between the varieties of modalities. The feature projectors were more tested by imposing triplet constraints so it can be made sure that representation shown by all items is used to minimise the overall gap using the semantic labels. It was also used to maximise the given distance of texts and images, which are semantic in nature. When joint exploitation is made, then it was found that the data preservation for the cross-modal semantic structure given for the multimedia data. It was evident in the end that four important benchmarks were used, and results taken from these benchmarks revealed the fact that the proposed method of ACMR is way better in an effective subspace representation with its effectiveness. It was also found that cross-modal retrieval methods were outperformed by the ACMR method by showing more superiority and effectiveness. This model can be further used and tested to get more opportunities in the future (Wang, Yang, Xu, Hanjalic, & Shen).

Article 2: A Thin-Plate Spline Calibration Model for Fingerprint Sensor Interoperability

            In this article, the focus of the authors is on the essence of biometric technology to see its usage and effectiveness in fingerprinting. The biometric technology is used by using a variety of tools to identify an individual or recognize him by looking at given features, which were earlier stored in the database of a biometric machine. It is a fact that regardless of various available biometric methods, fingerprinting is the widely used method. The authors have focused on the sensor interoperability of biometric, which is a system’s ability to do compensation of the variability, which has been shown by an individual’s biometric data. It has been observed that different kinds of biometric methods such as speech, face, and fingerprint have been associated with the intersensor performance, which is deemed poor. It is a well-known fact when fingerprint technology is used, the variation is there in acquiring the fingerprint images, and this variation happens because of sensing technology, scanning areas, as well as, sensor resolution. If a fingerprint is not able to deal with these variations in given circumstances, and its inability is confirmed, then the intersensor matching performance happens to be inferior. When such kinds of errors are shown by the fingerprint machine, then results will always be faulty to recognise fingerprints of individuals. So, it is important to come up with a framework or scheme of work, which deals with these variations in most of the situations. In this research work, by using the basis of the Thin-Plate Spline model, a nonlinear calibration scheme has been used, where a pair of fingerprint sensors were registered.

            It is important to know that authors have made sure that they use a model, which is good enough to deal with the fingerprint interoperability model. That’s why the calibration technique proposed in this research has been based on evidence provided by pairs of few images, which were acquired after the usage of two sensors. It was done so that a deformation model is generated to define the relationship upheld by two sensors, and this relationship is spatial as well. After implementing the methods, when data were analysed, it was found that the proposed model/method has proved effective to deal with the variations of intersensor geometric. It was observed that the performance of the biometric method was significantly improved. The sensor with this proposed model performed better than the other sensor. This calibration technique is a good indication that such kind of calibration models can also be used in the future, and more opportunities can be found in this regard. The authors also discussed the limitation of the study, which revealed that when there were sensors having distortions or different resolutions; the proposed scheme was not able to compensate for the given variations. But still, it is effective enough, and more research can be done to explore its further points, which can be beneficial for the biometric field (Ross & Nadgir, 2008)

Bibliography of Adversarial Cross-Modal Retrieval

Ross, A., & Nadgir, R. (2008). A Thin-Plate Spline Calibration Model for Fingerprint Sensor Interoperability. IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 20(8).

Wang, B., Yang, Y., Xu, X., Hanjalic, A., & Shen, H. T. (n.d.). Adversarial Cross-Modal Retrieval. MM’17, 154-162.