Machine Learning Based Miscellaneous Objects Detection with Application to Cancer Images

With the technological developments in artificial intelligence, traditional machine learning methods, and recently Deep Convolutional Neural networks (DCNNs) have accomplished great success in several vision related tasks. For example, object detection, object classification or recognition, and image segmentation are effectively achieved with the aid of aforementioned technologies. To train an effective DCNN model generally needs a big amount of diverse and balanced data. However, the DCNNs are quite often constrained by several aspects, such as (i) data annotation cost, (ii) long-tailed dispersal of data or imbalanced data, and (iii) scarcity of relevant data. Moreover, these days Generative Adversarial Networks (GANs) are widely used in image generation and synthesis related tasks. Object detection, being an important but challenging problem in the field of image processing and computer vision, plays an important role for several applications. While different methods exist to detect objects that appear in an image, a detailed analysis regarding common object detection is still lacking. This chapter pertains to detect objects that appear in an image with complex backgrounds using the Adaptive Boosting Algorithm (ABA). Different from several previously published chapters or research articles that focus on detection of six different objects, such as ships, vehicles, face, or eyes on standard datasets, this chapter focuses on real life objects, which include: (i) Melanoma (actual Malignant Melanoma, Nevus Spilus, and Blue Navy), (ii) license plates, (iii) vehicles, (iv) pedestrians, (v) players, and (vi) football. The ABA is applied on 8525 test images, which include: 251 Melanoma, 2512 license plates, 2400 vehicles, 2502 pedestrians, 590 players/sportsmen, and 270 football images. The chapter also analyses the challenges of current study during investigating real life several objects’ images and proposes a promising research direction, namely supervised learning cloud-based object detection. We are optimistic that this study will be useful for the researchers, practitioners, and beginners as a quick review with better understanding about object detection.