Significant Enhancement of Object Recognition Efficiency Using Human Cognition based Decision Clustering

Abstract

It is well known that human can recognize object-pattern better using its temporal description. In this paper both theoretical study and experiments were performed to translate this cognition principle into mathematical formula. In the implementation phase we considered breaking up of temporal data of human face into an assembly of time series data for each of which we obtained a decision as output of the chosen classifier. An assembly of decisions was thus resulted for a single temporal input data which was further judiciously clustered to obtain the final decision. Interestingly, the work also showed that the successive order of the time series data was not needed to be maintained; rather an assembly of randomly chosen multiple test data was important to obtain quite significant level of enhancement of classification accuracy. Thus, it gives new interpretation on temporal data based human cognition. The work also indicated that augmentation of this method with any classifier including those which used decision clustering tree, might yield quite a significant enhancement of recognition efficiency.

A Novel Approach for Object Recognition Using Decision Tree Clustering by Incorporating Multi-Level BPNN Classifiers and Hybrid Texture Features

“…It was assumed that the center of the largest cluster would provide the correct class decision and that the remaining clusters might contain incorrect data (methodology is also known). Clustering decisions, as per Kumar & Lahiri (2013b). In this work, it was assumed that correct test data might contain some noisy or incorrect data once it was discovered that, in accordance with traditional practice, the core TFMLBPNN-DTC based classifier accuracy on single test input was more than 80%.…”

“…Intensity level based multi fractal dimension (ILMFD), another cognition-inspired feature extraction technique, was also used in this work (the methodology to compute ILMFD is given in section 2.3.2.1). Fractal dimension (FD) was a tool that many researchers used to extract features from images for a variety of image processing applications (Kumar & Lahiri, 2013b;Tripathi et al, 2022). A FD is a pattern index or measure (in the form of integers or fractions) that is specifically used to count the complexity of objects' fractal patterns or feature sets.…”

“…It can identify a wide range of picture classes and effectively describe the intrinsic information of the objects, something that other types of texture analysis methods are unable to accomplish. It was suggested that ILMFD be used to classify and identify real-world objects and images when the ROI requires intricate analysis (Kumar & Lahiri, 2013b). Their findings suggested that it was a superior choice over the conventional intensity-based bit-plane slicing (n-bit slicing).…”

An Efficient and Non-Invasive Approach to Identify and Classify COVID-19 from CT Scan Images using Machine Learning in Combination with Hybrid Texture Features

“…This is the set of "n" fractal dimensions, as seen in Fig. 3, which correspond to all "n" slices of an image based on equal intensity intervals [47][48] and were ascertained using the boxcounting approach [46]. Box counting is a method for gathering information for analyzing intricate patterns by breaking an image into progressively smaller parts, usually the shape of a "box," then analyzing the pieces at each lower scale.…”

Analysis of Methods of Machine Learning Techniques for Detection and Classification of Brown Spot (Rice) Disease

“…The same algorithm can be replicated in the case of many other staple crops with a very minor change. The use of Intensity level-based multi-fractal dimension (ILMFD) [20] with ANN (Artificial Neural Network) [21] can also be done and a parallel system for disease identification can be developed.…”