Methods of image pre-processing for automated object recognition systems

This study developed a new method of image pre-processing for use in automated object recognition systems, allowing for increased accuracy, reliability, and speed of recognition under varying lighting conditions and other external influences. The proposed method is based on contrast-limited adaptive histogram equalization (CLAHE) and gamma correction. To enhance illumination, the method employs a novel adaptive cutoff point to equalize the histogram based on the texture information of image blocks. To address issues such as halo artifacts, under-gain, and over-gain, a first gamma correction function is proposed for integration into the CLAHE structure, which enhances the overall variation of gray shades in the image. A second gamma correction function is proposed to mitigate the negative impact of varying illumination on dark and light regions. For the development of a new image pre-processing method based on a convolutional neural network, a unique layer arrangement is proposed, featuring several layers of convolution, activation, pooling, and normalization. Experimental tests have demonstrated the effectiveness of the developed method, which achieved an accuracy coefficient of 93.74%. This was evaluated by comparing its performance with other image pre-processing methods within an automated face recognition system based on Local Binary Patterns (LBP) and the Support Vector Machine (SVM). The results confirmed that the developed method can improve existing object recognition systems or serve as a foundation for creating new automated software applications requiring high reliability in image recognition.