[en] The LHC (Large Hadron Collider) provides proton-proton collisions to CERN (European Organization for Nuclear Research) experiments at a 7 TeV center of mass energy since March 2010. The LHC has been designed in particular to allow the Higgs boson searches, particle predicted in the standard model but still not discovered until today, in the whole mass range where it is expected. This work is a contribution to the Higgs boson searches in CMS (Compact Muon Solenoid), one of the four big detectors at LHC. The thesis develops several tools which allow to measure the backgrounds and to improve the discovery potential. A new tool for recovery of photons emitted by leptons in the final state: H → ZZ(*) → 4l (l=e, μ) has been developed in this thesis. This method that recovers a variable number of photons per event, performs better than the method previously used in CMS and improves Z⁰ and Higgs boson mass resolution. A 5% gain on the significance to observe a Higgs boson in this channel is reached. The second part of this work deals with studies of the backgrounds and the search for a light Higgs boson (110 < m_H < 140 GeV) in the channel H → γγ. A new tool for γ/π⁰ discrimination with a neural network has been developed to reject photons coming from π⁰ decays, copiously produced in QCD jets. The neural network performance is examined in details. The neural network is then used as 'template' variable to measure γ+X process in data with 10 nb^-1 of integrated luminosity. The measurement of γγ+X process is also prepared with simulation in the hypothesis of a 10 pb^-1 luminosity. Taking into account higher order kinematic effects is necessary to perform the best prediction of H → γγ signal and backgrounds. In the thesis this is carried out with a reweighing method, at NNLO for gg → H → γγ process and for the first time at NLO for γγ+X process, in both cases with doubly differential distributions. Reweighing procedure and γ/π⁰ neural network are then integrated in the H → γγ analysis to improve CMS sensitivity in the standard model and beyond. (author)