Highly Contrast Image Correction for Dim Boundary Separation of Image Semantic Segmentation

The efficiency and accuracy of the image semantic segmentation algorithm represent a trade-off relationship, loss tends to increase as model structure simplifies improve efficiency. Developing more efficient accurate algorithms requires methods complement them. In this study, we applied logarithmic-exponential mixture (LEM) function for gamma correction images segmentation. basic used in work was produced by constructing full convolution neural network based on MobileNetV2. To avoid noise input compression, corrected training validation with from 1/8 8 (7 different levels) before doing convolution. We evaluated models using Tensorflow deep-learning library Python. compared LEM conventional function. prediction masks proposed had relatively small fluctuations upon change. For that have shadows overlapped object, object better distinguished values. dark images, effective. results indicated could unclear edges. believe presented will guide further studies improvement recognition applicable future devices, such autonomous vehicles mobile robots.