Review of Image Recognition from Image Repository

Abstract

The work mainly focuses on the need of building a hybrid model by combining the different available method and techniques used for features extraction, detection, and matching. The enhanced model developed is used to identify the objects based on shape. The features are extracted from universally accepted images considered as candidate object. Number of works on image extraction has pointed out the difficulties in extracting the needed image. The paper highlights the shortfalls and the need to address the shortfalls. The review conducted provides the research gap in the industry and various directions the work is heading.

Introduction

I. INTRODUCTION

The most important step in image retrieval is to find the correct features of the candidate object. The development of the machine learning based feature detection methods is required to fulfil the growing need of object detection. To build a novel model for the feature detection, classification, and matching is needed which can perform better than previously defined models. The easier way of implementing the complex image processing methods using Python is also the major point of interest.

The review of the most popular current methods in the industry has reached reasoning that not one feature extraction and detection method is universally acceptable. The fact that each image has a different set of attributes which defines it uniqueness leads to the downfall of the developed model which would have provided wonderful results for a different image. The framework and the algorithms can be used a basis for building various software which are used for general purposes as more number of different type of input images can be predicted [2].

II. ANALYSIS AND DEVELOPMENT OF IMAGE RETRIEVAL SYSTEM

The present world trends at fast and accurate data retrieval from almost all visible elements. The focus is shifted towards the images as images have lot of information hidden in them naturally and is still not extracted as efficiently as it is done by humans by computers. There is a large gamete of data residing in classified images like the business images, satellite images, medical images and so on. The analysis of any image needs appropriate data to be gathered from the image before it is processed. There are a number of reasons for faulty data extraction due to the vast number of categories available and each image being a completely different unit all by itself. This leads to incomplete information and to further for inappropriate conclusion.

Number of researchers worked on image mining. There are a number of different and efficient methods developed to acquire and store these images in the recent times. This prompts for huge image databases holding important information in them. Image mining extracts associations among mages, patters of interest, hidden information, etc. This research involves machine learning, image processing, database, computer vision, AI, image retrieval and data/image mining

III. CLASSIFICATION OF IMAGES

Image analysis can be broadly classified into two types as pixel based and object based. Pixel based classification methods are commonly used in applications classifying remote sensed images, object based images classification has shown lot of potential in the coming years including the area of remote sensing [13]. There are a number of comparative studies to examine these two classification methods. The application in these methods are used seem to determine the accuracy and the aptness of use. Though pixel based and object based classifications are efficient in specific environments, the latter performs better when compared for overall classification on remotely sensed images. There have a number of studies to bridge the gap between pixel based and object based classification with the help of machine learning algorithms like the Random forest, Decision tree and Support Vector machines[7].

IV. IMPACT OF MACHINE LEARNING ON IMAGE RETRIEVAL

The use of powerful and robust machine learning languages have made way for more impactful research in the area of image retrieval from moving as well as still images. There a number of Content Based Image Retrieval (CBIR) prototypes that are available commercially and are in good use. CBIR targets scanning image databases for explicit images that are like a given question image. It likewise centers at growing new strategies that help powerful looking and perusing of huge digital image libraries in view of consequently determined imagery highlights. It is a quickly extending research region arranged at the convergence of databases, data recovery, and computer vision. Despite the fact that CBIR is as yet young, there has been huge amount of earlier work.

The CBIR concentrates on Image 'features' to enable the query and have been the new focal point of investigations of image databases. The features further can be delegated as low-level and significant level features. Clients can query example images in light of these features like surface, variety, shape, district and others. By closeness correlation the objective image from the image storehouse is recovered. In the mean time, the following significant stage is centered around grouping strategies[5]. Clustering algorithms can offer predominant association of multidimensional information for successful retrieval. Clustering algorithms permit a closest neighbor search to be proficiently performed. Consequently, the image mining is quickly acquiring consideration among the specialists in the field of information mining, data retrieval and multimedia databases. Spatial Databases is one of the ideas which assumes a significant part in Multimedia System. Semantically meaningful data can be extracted by research from images.

Number of analysts chipped away at image mining. There are various unique and productive strategies created to gain and store these images in the new times. This prompts for colossal image databases holding significant information in them [3]. Image mining gets relationship among images, patters of interest, hidden information, and so forth this area of exploration of multidisciplinary effect that makes way for number of applications of AI, processing the candidate image, retrieval of image, computer vision, database and AI [8]. Provided with a content based tissue image mining on Biological data and then mining the needed areas for analysis and further research. The work suggests high level high quality images from Tissue Microarray Analysis (TMA) is considered as input, this image is filed, put away and used for mining based on content. Large data into information and good performance are two features of TMA systems[10].

A wavelet transforms technique for image mining. It uses identical patterns to identify and mine correct scene that needs to be related to a specific class, aiding different expectation and forecasting mechanisms. It is a three-venture technique for example image gathering, learning and classification[11].

V. DATA MINING APPROACH

One of the papers proposed a mining approach of images using clustering and data processing. This technique was later used for satellite images of clouds to specifically forecast weather conditions. Image capturing frequency is defined as one image in the beginning and then next image every one hour based on the climatic conditions. In case of turbulence, the frequency may be increased to improve accuracy of forecast.

This leads to huge collection of data/images with large high definition files[4]. Transmission with extreme durability is a requirement. In their methodology, data mining clustering technique along with Vector Quantization (VQ) is implemented to cluster and compact static color image.

A direction to the image mining: the standard apparatus, subjects, system, application of analysis to such captured clinical images. The paper provides technique for data mining in picture document system. Image description was considered the efficient way and a most common way and reasonable way for classification and identification of candidate image. The image is described as symbolic image and knowledge-engineering is applied as a strategy to identify the attributes [6]. The specialized mechanism identifies the descriptions of the images called the features. Digital-image processing can be implemented to acquire better imaging of explicit image attributes, or to get master autonomous trademark assessment.

There are various techniques of image mining to provide data for MRI images. MRI was conducted for the brain of a human being. Segmentation of the brain images assumes an essential part in a few clinical imaging applications by computerizing or helping the depiction of anatomical arrangements and extra locales of interest. Automatic recognition of tumors in a few clinical images is empowered by the necessity of better exactness while taking care of with a human existence [12]. Additionally, the computer help is requested in clinical organizations attributable to the truth that it conceivably will advance the aftereffects of people in such a domain where the bogus pessimistic cases should be at an extremely low rate [9].

Conclusion

The paper provides the review of work so far on image recognition and the methods utilized for this purpose. The different areas and applications where image retrieval and recognition is involves is also under consideration. Content based image retrieval with help of machine learning techniques being the latest in the industry due its accuracy is yet to pick up speed and more accuracy. The tradeoff between the image features needed and the accuracy and speed are still a matter of concern. Overfitting and under fitting of data is still a matter of debate for each case and hence generalization is a distinct reality for machine learning. Satellite images for climate prediction, medical images of TMA and MRI are also considered in this paper. There is major gap to be bridged between the need of the industry and the existing present.

References

[1] T. Blaschke, Object based image analysis for remote sensing, ISPRS Journal of Photogrammetry and Remote Sensing, Volume 65, Issue 1, 2010, Pages 2-16, ISSN 0924-2716, https://doi.org/10.1016/j.isprsjprs.2009.06.004 [2] Yu Zhao, Jingyu Wang, Qi Qi. \"MindCamera: Interactive Image Retrieval and Synthesis\", 2018 IEEE 13th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP), 2018 [3] Shixing Chen, Ming Dong, Dongxiao Zhu. \"Learning and Interpreting Features to Rank\", International Journal of Multimedia Data Engineering and Management, 2018. [4] Shixing Chen, Caojin Zhang, Ming Dong. \"Deep Age Estimation: From Classification to Ranking\", IEEE Transactions on Multimedia, 2017. [5] Yu Wang, Yang Feng, Haofu Liao, Jiebo Luo, Xiangyang Xu. \"Do They All Look the Same? Deciphering Chinese, Japanese and Korean by Fine-Grained Deep Learning\", 2018 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR), 2018. [6] Mayank Meghawat, Satyendra Yadav, Debanjan Mahata, Yifang Yin, Rajiv Ratn Shah, Roger Zimmermann. \"A Multimodal Approach to Predict Social Media Popularity\", 2018 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR), 2018. [7] Zechao Li, Jinhui Tang, Tao Mei. \"Deep Collaborative Embedding for Social Image Understanding\", IEEE Transactions on Pattern, Analysis and Machine Intelligence, 2019. [8] Jingyu Wang, Yu Zhao, Qi Qi, Qiming Huo, Jian Zou, Ce Ge, Jianxin Liao. \"MindCamera: Interactive Sketch-Based Image Retrieval and Synthesis\", IEEE Access, 2018 [9] E.H Zhang, Y. C. Li, J. H. Duan. \"Moving object detection based on confidence factor and CSLBP features\", The Imaging Science Journal, 2016 [10] Bankhead, P., Loughrey, M.B., Fernández, J.A. et al. QuPath: Open source software for digital pathology image analysis. Sci Rep 7, 16878 (2017). https://doi.org/10.1038/s41598-017-17204-5 [11] Prasad, L., & Iyengar, S.S. (1997). WAVELET ANALYSIS with Applications to IMAGE PROCESSING (1st ed.). CRC Press. https://doi.org/10.1201/9780367811310 [12] Jose Bernal, Kaisar Kushibar, Daniel S. Asfaw, Sergi Valverde, Arnau Oliver, Robert Martí, Xavier Lladó,. Deep convolutional neural networks for brain image analysis on magnetic resonance imaging: a review, Artificial Intelligence in Medicine, Volume 95, 2019, Pages 64-81. [13] T. Blaschke, Object based image analysis for remote sensing, ISPRS Journal of Photogrammetry and Remote Sensing, Volume 65, Issue 1, 2010, Pages 2-16

Copyright

Copyright © 2022 Manish Kumar Thakur, Raju Ramakrishna Gondkar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.