Deep Learning Algorithms-based Object Detection and Localization Revisited

Abstract

The computer vision (CV) is an emerging area with sundry promises. This communication encompasses the past development, recent trends and future directions of the CV in the context of deep learning (DL) algorithms-based object detections and localizations techniques. To identify the object location inside an image and recognize it by a computer program as fast as the human brain the machine learning and DL techniques have been evolved. However, the main limitations of the machine are related to the prolonged time consumption to handle vast amount of data to perform the same task as the human brain. To overcome these shortcomings, the convolution neural networks (NNs)-based deep NN has been developed, which detects and classifies the object with high precision. To train the deep NNs, massive amount of data (in the form of images and videos) and time is needed, making the computational cost of the CV very high. Thus, transfer learning techniques have been proposed wherein a model trained on one task can be reused on another linked task, thereby producing excellent outcomes. In this spirit, diverse DL-based algorithms have been introduced to detect and classify the object. These algorithms include the region-based convolutional NN (R-CNN), fast R-CNN, Faster R-CNN, mask E-CNN and You Only Look Once. A comparative evaluation among these techniques has been made to reveal their merits and demerits in the CV.

Acute lymphoblastic leukemia image segmentation based on modified HSV model

“…Several image segmentation methods have been proposed, including deep learning [14], Fuzzy C-means [15], K-means clustering, Otsu's thresholding [16], Watershed [17], deep learning, and other algorithms [18]. For example, in [19], the authors proposed a new segmentation approach using the k-means cluster and morphological operation on 642 images and got an overall segmentation accuracy of 97.47.…”

Melanoma Skin Cancer Classification based on CNN Deep Learning Algorithms

“…Algorithms are being programmed onto computers to do this. Low-complexity approaches are developed to aid laypeople and are optimized for use on mobile devices like tablets and smartphones [4,5]. Expert judgment in this area, however, necessitates the use of complex computational methods and apparatus [6].…”

A novel medical image enhancement technique based on hybrid method

“…Consequently, image enhancement is an essential part of improving the visual quality of computer vision, pattern recognition, and the processing of digital images, and it plays a significant role in all three of these areas. Image enhancement is used to increase the contrast of an image by sharpening the edge pixels concerning the background pixels [21]. As a result, it reduces background noise and enlarges individual pixels so that important details can be picked out of an image-applications of image enhancement range from medical image analysis to remotely sensed imagery and beyond [22].…”