Reinforced Contrast Adaptation

Abstract

Traditional image enhancement algorithms do not account for the subjective evaluation of human operators. Every observer has a different opinion of an ideally enhanced image. Automated Techniques for obtaining a subjectively ideal image enhancement are desirable, but currently do not exist. In this paper, we demonstrate that Reinforcement Learning is a potential method for solving this problem. We have developed an agent that uses the Q-learning algorithm. The agent modifies the contrast of an image with a simple linear point transformation based on the histogram of the image and feedback it receives from human observers. The results of several testing sessions have indicated that the agent performs well within a limited number of iterations.

Automatic Selection of Object Recognition Methods Using Reinforcement Learning

“…The parameters learned by RL were shown to be superior to the parameters previously recommended. Other applications of RL to learn parameters of image segmentation algorithms includes: contrast adaptation (Tizhoosh and Taylor, 2006), finding the appropriate threshold in order to convert an image to a binary one (Yin, 2002;Shokri and Tizhoosh, 2003Sahba et al, 2008) and detection of patterns in satellite images (Hossain et al, 1999).…”

A reinforcement agent for object segmentation in ultrasound images

“…Recently, some other approaches have been introduced which show the application of RL on some image-based problems (Sahba & Tizhoosh, 2003;Sahba, Tizhoosh, & Salama, 2005a, Sahba, Tizhoosh, & Salama, 2006a, Sahba, Tizhoosh, & Salama, 2006bShokri & Tizhoosh, 2004;Taylor, 2006;Tizhoosh & Taylor, 2006;Yin, 2002).…”

A reinforcement learning framework for parameter control in computer vision applications

“…Several various methods of defining state are seen in the literature with respect to image-related problems. One can use raw image data [4] or histogram statistics [11] but this approach can generally lead to extremely large state spaces, which are undesirable. One may use local statistics, or even subjective evaluation of the image.…”

“…The evaluation can either be subjective or objective. The former, however, as performed in [11] requires the participation of human subjects which can make experimentation lengthy and expensive. On the other hand it is desirable in certain algorithms, allowing for observer-dependent vision tasks.…”

“…Historically, the literature has not offered many examples where image-based problems have been solved using this family of learning algorithms, but recently, research has shown that RL can indeed be applied to such problems [11,9,13]. In this paper, we explore some of the obstacles faced when integrating RL agents into imagebased applications and discuss how they might be overcome (Section 2).…”